EMPLOYING INSECT ANTENNAL LOBE OLFACTORY
NEURAL SIGNALS FOR NON-INVASIVE DISEASE DETECTION
By
Alexander J. Farnum
A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Biomedical Engineering — Doctor of Philosophy
2023
ABSTRACT
Gas-based chemical sensors have proven invaluable for investigating the underlying
chemical configuration of a particular odorant. Technologies such as gas chromatography-mass
spectrometry and hyphenated ion mobility spectrometry have elucidated numerous causal
factors and biochemical pathways contributing to a variety of healthy and pathologic conditions.
Nevertheless, these technologies face innate challenges that have precluded their adoption and
implementation in the clinical environment. As a result, a considerable amount of research and
development has been geared towards fabricating cheap, easy-to-use, and highly portable
electronic noses. These devices have demonstrated excellent potential for applications such as
environmental monitoring, food analysis, and forensic science. Moreover, innovations in
nanotechnology and other materials science fields have spurred the ideation and creation of
highly efficient electronic noses. However, the broad range and low concentrations of chemical
metabolites observed in breath profiles, hinders their use as medical diagnostics for complex
diseases. Here, this work proposes a novel technology utilizing biologically based chemical
biosensors to accurately characterize the volatile profiles associated with pathological disease
states, especially that of cancer. The development of this powerful and efficient gas-sensing
system has the potential for use in a variety of real-world contexts, including homeland security,
law enforcement, and medicine.
It is well known that the presence of disease alters underlying biochemical processes,
thereby influencing metabolic byproducts and the volatile chemicals excreted via the breath.
Existing manmade chemical sensors as medical diagnostics lack the ability to differentiate the
breath profiles of healthy individuals from those with complex diseases. Chapter 1 investigates
the field of volatolomics as a whole, including the gas-based identification of ‘simple’ diseases
and the application of state-of-the-art sensor technologies to diagnose complex pathologies. In
chapter 2, the methodology for all experiments involving this novel gas-based biosensor in the
context of disease diagnosis is discussed. Chapters 3, 4, and 5 detail applied research which
validates the biosensors’ powerful abilities to differentiate chemicals and chemical profiles. This
work serves to establish its potential as a non-invasive medical diagnostic using biological
matrices, such as breath profiles. Finally, chapter 6 discusses the current limitations of the proof-
of-concept technology as well as modifications that will mitigate such limitations and aid in the
creation of an effective state-of-the-art breath-based medical diagnostic.
TABLE OF CONTENTS
CHAPTER 1 | INTRODUCTION ........................................................................................................ 1
Volatolomics............................................................................................................................. 1
Selective Gaseous Sensors ....................................................................................................... 4
Gas Chromatography-Mass Spectrometry (GC-MS)................................................................. 8
Ion Mobility Spectrometry (IMS) & High-Field Asymmetric Waveform Ion Mobility
Spectrometry (FAIMS) ............................................................................................................ 12
Electronic Noses ..................................................................................................................... 17
Biological Olfaction ................................................................................................................ 23
Biosensors & Bioelectronic Noses .......................................................................................... 31
Forward Engineering Novel Insect Biosensors ....................................................................... 37
CHAPTER 2 | METHODOLOGY ...................................................................................................... 39
Odorant Delivery .................................................................................................................... 39
Signal Recording & Electrode Preparation ............................................................................. 43
Stimulus Creation ................................................................................................................... 51
Husbandry .............................................................................................................................. 55
Surgery & Electrophysiology .................................................................................................. 59
Signal Analysis ........................................................................................................................ 65
CHAPTER 3 | DISCRIMINATING CANCER BIOMARKERS USING A NOVEL INSECT BIOSENSOR ..... 71
Chemical Sensors ................................................................................................................... 71
Cellular Metabolism & Breath-Based Biomarkers .................................................................. 72
Locust Olfaction ..................................................................................................................... 75
Locust-Based Cancer Biomarker Differentiation .................................................................... 78
Outlook .................................................................................................................................. 93
CHAPTER 4 | CANCER CELL LINE DIFFERENTIATION BY AN INSECT CHEMICAL BIOSENSOR ........ 95
Cancer Metabolism ................................................................................................................ 95
Volatile-Based Cancer Detection............................................................................................ 96
Locust-Based Cancer Biosensor ............................................................................................. 99
Oral Cancer Classification via Multi-Dimensional Neural Signal Analysis............................. 103
Time-Matched Cancer Volatile Discrimination .................................................................... 111
Rapid Detection and Identification of Complex Cancer Volatile Headspace ....................... 115
Outlook ................................................................................................................................ 121
Acknowledgments ................................................................................................................ 124
iv
CHAPTER 5 | DISCRIMINATING CANCER BIOMARKERS USING HONEYBEE NEURAL
RESPONSES ................................................................................................................................. 125
Application-Specific Chemical Sensors ................................................................................. 125
Honeybee Olfaction ............................................................................................................. 127
Honeybee-Based Cancer Biomarker Differentiation ............................................................ 129
Outlook ................................................................................................................................ 133
Acknowledgments ................................................................................................................ 134
CHAPTER 6 | ONGOING WORK AND FUTURE DIRECTIONS ........................................................ 135
Introduction ......................................................................................................................... 135
Current Limitations .............................................................................................................. 136
Multi-Electrode Array Modifications.................................................................................... 138
Sensor Calibration ................................................................................................................ 146
Brain-on-a-Chip .................................................................................................................... 148
In Vivo Diagnostic Validation................................................................................................ 150
Outlook ................................................................................................................................ 150
REFERENCES ............................................................................................................................... 151
v
CHAPTER 1 | INTRODUCTION
Volatolomics
Chemical molecules are the building blocks of the material world. Depending on their physical
properties and environmental conditions, molecules can exist in solid, liquid, or gaseous forms.
For example, a sufficient increase in temperature can break the covalent bonds of H2O molecules,
evaporating the hydrogen and oxygen atoms into their respective gases. The resultant volatile
chemicals can convey a considerable amount of information about the underlying chemical
composition of the stimulus itself. Since atoms in the gaseous state can diffuse more freely than
those of solids or liquids, gas-based sensing offers a method for non-invasive chemical analysis,
particularly important for toxic, delicate, or otherwise inaccessible stimuli. Today, such gas
sensors exist in a litany of fields, from law enforcement to environmental monitoring to medicine.
However, given the millions of known chemicals and minute conformational differences between
individual species, accurately characterizing a volatile requires exceptionally high precision. This
task is made even more difficult when one considers that the vast majority of stimuli are
constructed of heterogeneous molecular matrices, significantly complicating the associated
volatile profiles. These demands have led to the development and progression of a number of
disparate scientific domains. Volatolomics, one such field still in its infancy, seeks to apply
chemical sensing technologies intended to aid in the identification of the unique volatile organic
compounds (VOCs) emitted by biological organisms[1]. Volatolomics has direct implications for a
wide variety of societal sectors. In food science, the VOC profiles of specific foodstuffs have been
observed to indicate overall quality and possible nutritional deterioration[2]. In environmental
1
Disease Source Smell
Diabetic Ketoacidosis Breath Rotting apples, acetone
Renal Failure Breath Stale urine
Intestinal Obstruction Breath Feculent, foul
Liver failure Breath Musty fish, raw liver, feculent
Fetor hepaticus Breath Newly mown clover, sweet
Pneumonia Breath Putrid
Typhoid Skin Freshly baked brown bread
Yellow Fever Skin Butcher’s shop
Squamous-cell carcinoma Skin Offensive odor
Diphtheria Sweat Sweet
Rubella Sweat Freshly plucked feathers
Schizophrenia Sweat Mildly acetic
Bladder Infection Urine Ammonia
Table 1.1 | Characteristic Odors of Disease. Patient odor has long been used by physicians as a
diagnostic element. A large number of diseases have been documented to elicit stereotypical
odors, perceivable even with the human nose, which displays relatively poor olfactory
capabilities relative to other animals. Adapted from [3, 4].
maintenance, the growing number of densely populated cities has led to an increased demand
to actively monitor air pollution levels via IoT (Internet of Things) devices[5]. In homeland
security, detecting and differentiating between the volatile profiles of various explosive
precursors could allow means for highly accurate, non-destructive sampling and safer public
spaces[6].
Certain diseases have been known to exude particular odors since antiquity[4, 7-11] and
physicians were taught to consider smell as a key factor for assessing a patient’s health until
relatively recently (Table 1.1) [12-14]. Nevertheless, this subjective olfactory-based diagnostic
fell into disuse with the advent of objective technologically based diagnostics. In fact, gas-sensors
as medical devices are just now becoming standard tools in research laboratories and clinical
practice. Capnography, the monitoring of CO2 levels in patients’ exhaled breath, has enabled
2
Figure 1.1 | Real time monitoring of CO2 using a handheld capnograph device. Capnography
provides insight into the amount of expired carbon dioxide in exhaled breath. As such, it can aid
in breath sampling protocols by standardizing collection procedures using chemically derived
information. Reproduced from [15].
anesthesiologists to significantly mitigate hypoxia induction and is useful for enhancing the
quality of breath samples (Figure 1.1) [16, 17]. For individuals with suspected bronchial asthma
a) b)
Figure 1.2 | Breath profiles include various volatile metabolites useful for disease diagnosis.
Exhaled breath contains a myriad chemical information pertaining to underlying biochemical
processes. Analysis of breath profile constituents can aid in non-invasive disease diagnosis. a)
Metabolic byproducts from cells and tissues are passed into the blood stream based on the
partition coefficient between fat and blood (lf:b). As the pulmonary circulatory system transports
blood and metabolic byproducts to the lungs, molecules diffuse into the alveoli of the lungs based
on the partition coefficient between blood and air (lb:a). Reproduced from [18]. b) These volatile
biomarkers are excreted via the breath in trace-level concentrations, which can aid in identifying
underlying metabolic processes. Reproduced from [19].
3
or other inflammatory airway diseases, fractional exhaled nitric oxide (FENO) measurements are
commonplace[20-22] and may even be superior to conventional diagnostic approaches[23].
Exhaled levels of acetone have been shown to increase significantly in those with type I diabetes,
especially during bouts of ketoacidosis[24-29]. Even nonpathogenic, yet suboptimal cellular
processes, such as free radical formation and systemic oxidative stress can be determined by
elevated concentrations of ethane, pentane, and other unbranched hydrocarbons (Figure 1.2)
[30-33].
Selective Gaseous Sensors
Independent of the specific analytical technology, all gaseous chemical sensors operate according
to two basic principles: chemical recognition and signal transduction. Initially, the device must
incorporate methods for interacting with the analytes of interest, oftentimes separating out
individual chemicals based on one or more physicochemical molecular properties. Thereafter, a
readout signal is generated permitting direct stimulus classification or subsequent analysis
designed to aid in stimulus classification. Two approaches to gas sensing have been realized due
to the implementation of novel technologies and new knowledge acquisition. Targeted bottom-
up approaches first identify volatile biomarkers corresponding to a disease of interest and
subsequently look to fabricate gas sensing devices most equipped to separate out and identify
these chemicals. Conversely, untargeted top-down strategies utilize broad-scale sensors that aim
to characterize highly informative features present within a sample with no preconception of
potential biomarker identities[1].
4
a) b)
Figure 1.3 | Single and multi-component sensor schematics. a) Sensors highly specific to
individual chemicals can provide exceptional sensitivity owing to the sheer number of integrated
sensors. Higher sensor concentrations enable a higher rate of analyte-sensor chemical reactions,
capable of eliciting high signal-to-noise ratios. b) Multi-component sensors are necessary for
highly heterogeneous gaseous stimuli. Although these sensors do not display the fine detection
limits of single-component sensors, they can be made responsive to a range of chemical species
whose identity is dependent on incorporated sensor materials.
For pathophysiological conditions well characterized by an individual chemical species,
targeted sensors can be highly effective for classification (Figure 1.3a). These selective sensor
technologies ought to be able to detect the volatile marker of interest at the appropriate
concentration. Since exhaled breath aerosol consists predominantly of a mixture of nitrogen,
oxygen, carbon dioxide, water vapor and inert gases, devices such as capnographs do not face
significantly high sensitivity requirements[7]. However, in addition to these main breath
constituents, more than 3400 other volatile biomarker species have been found at trace levels,
though some of these are likely of exogeneous origin[34-36]. Selective sensors are constructed
according to a lock-and-key type model, in which an individual chemical species and
corresponding sensor interface are highly specific for one another. This ligand specificity enables
high degrees of sensitivity, capable of achieving the necessary parts-per-million and parts-per-
billion detection thresholds for some of these compounds. Nitric oxide, for example, is routinely
present at low parts-per-billion levels in healthy individuals, yet average concentrations in those
5
with bronchial asthma are increased three-fold[37-41]. Innovative functional transducer surface
modifications, such as electrode platinization and electrochemical film layering, have been found
to heighten nitric oxide sensitivity and selectivity, culminating in more robust patient
classification[42-45]. Alternatively, exhaled acetone concentrations for healthy individuals have
been noted to be between 300 and 900 parts-per-billion, whereas diabetics’ levels often exceed
1.8 parts-per-million[46, 47]. During bouts of diabetic ketoacidosis, acetone concentrations can
reach a thousand parts-per-million or higher (Figure 1.4). Novel selective sensor designs have
continued to improve device sensitivity reaching low parts-per-billion ranges[48-54]. Such highly
selective sensors are simple, robust, and provide adequate discriminability for some diseases[55].
However, they do not offer sufficient diagnostic power for all, or even most pathophysiological
conditions[45, 47]. The presence of confounding factors (i.e. concomitant disease, smoking
a) b)
Figure 1.4 | Exhaled acetone concentrations are indicative of underlying ketogenic conditions.
a) The overproduction of ketone bodies induces changes in blood chemistry. Elevated levels of
breath-based acetone correspond to blood serum concentrations and have been observed for
those experiencing ketogenetic conditions and diabetics at large. Reproduced from [1]. b) A
portable chemo-resistive acetone sensor incorporating interdigitated platinum electrodes with a
resistance temperature detector (RTD) is shown. Back-heating by the platinum electrode
increases the sensitivity of silicon-doped tungsten oxide nanoparticles on the electrode surface
with detection limits of 20 parts-per-billion. Reproduced from [50].
6
status, etc.) or large variations in individual breath-based VOCs, can complicate real-world
diagnostic implementation[39, 41, 56-60]. The majority of diseases are complex phenomena that
have multifaceted consequences on cells, tissues, and entire organs. As such, the immediate
effects of a specific disorder can elicit downstream signaling cascades, increasing or decreasing
levels of various primary as well as secondary metabolites. Devices that aim for high chemical
specificity with narrowly tuned sensors are effective for a handful of diseases but are generally
not well-suited for multi-component identification.
Instead of relying on single-component sensing methods for disease detection and
classification, the majority of diseases have been found to be associated with altered levels of a
host of disparate VOCs[15]. This suggests that multi-component discrimination techniques may
offer superior diagnostic potential, especially for heterogeneous samples[61] (Figure 1.3b). The
breath profile of a particular individual is a snapshot in time reflecting a unique combination of
genetic, dietary, environmental and lifestyle influences. As such, there may not be a linear
correlation between any single biomarker and an underlying pathological state. That is, the
concentration of a particular VOC found in a healthy individual may be the same as that found in
a diseased patient. Even two patients expressing the same pathology will have some degree of
variability in VOC concentrations, based on the resolution of the analytical technology. Multi-
component sensors approach this challenge by integrating various sensor measurements to
glean relevant stimulus information from a large proportion of the constituent volatile chemicals.
Such systems are characterized by more degrees of freedom and superior discriminatory power
than selective systems. Even for diseases well-characterized by individual chemical species, broad
spectrum chemical analyses can offer superior diagnostic precision due to the idiosyncratic
7
nature of breath samples[36, 62]. A number of multi-component gas sensing technologies have
undergone rigorous laboratory testing with the goal of demonstrating sufficient discriminability
to warrant clinical implementation.
Gas Chromatography-Mass Spectrometry (GC-MS)
One untargeted technology that has shown tremendous potential for analyzing breath samples
is gas-chromatography mass spectrometry (GC-MS) (Figure 1.5). Gaseous samples are initially
collected and concentrated onto an analyte trapping device, usually in the form of a sorbent tube
or microextraction fiber. The matrix containing these adsorbed sample chemicals is then placed
inside the injection port of a gas chromatograph. Here, an ultra-pure carrier gas is passed through
the sample as the oven temperature is incrementally increased to facilitate analyte-sorbent
desorption. With continual heating, chemicals are desorbed in accordance with their volatility
before entering a capillary column—highly volatile components are desorbed at lower oven
temperatures followed by the elution of those with lower volatilities at higher temperatures.
Column- and molecule-specific properties determine the state phase probability and rate of
transmission through the column. The larger and less volatile components will display longer
column retention times and will elute later than smaller, highly volatile chemicals (Figure 1.5b).
Prior to pairing with mass spectrometers, detectors such as flame-ionization and thermal
conductivity detectors, were incorporated to measure the elution time of each molecule.
However, similar retention times for compounds, such as pentane and isoprene, have been
observed, confounding readout accuracy[63, 64]. The spatiotemporally stratified gas molecules
are depressurized at the GC-MS interface before entering the mass spectrometer. Once the
8
a)
b) c) d)
Figure 1.5 | GC-MS. a) GC-MS schematic illustrating separate components. Carrier gas is purified
and passes over the heated sample at the sample inlet/injection port. Molecular separation
occurs along the length of the capillary column. Spatiotemporally segregated analytes are
ionized, fragmented, and separated by mass-charge ratio in the mass spectrometer (MS). b) Gas
chromatography separates ions based on size and volatility, or gaseous vs. liquid state phase
probability. c) Electron ionization schematic. An electron beam interacts with molecules of a low-
pressure gas cloud. Proximal electrons energize the analyte molecule, cueing electron release
and formation of a positively charged ion. d) Different ionization techniques can result in
completely different fragmentation patterns, eliciting contrasting mass spectra. Amphetamine
(left) and a derivative of amphetamine (right) were both observed to exhibit variability in
fragmentation patterns based on the ionization technique. Reproduced from [65].
9
eluted chemicals have entered the mass spectrometer, they are bombarded with 70-eV
electrons, ionized, and, with sufficient energy, rapidly fragment to smaller ions. The low-pressure
environment of the mass spectrometer enables ion flow trajectories to be dictated purely by
electromagnetic properties, with little to no influence from ambient air particles. The stream of
ions travels along the mass-to-charge analyzer until striking a detector. An overall sample
chromatogram indicates the relative quantity of unique molecules separated by their retention
time. At each time step, a unique mass spectrum is constructed, detailing an analyte-specific
molecular configuration. The unique separation techniques utilized by gas chromatography and
mass spectrometry serves to significantly enhance molecular discriminability than using either
one as standalone sensors[66, 67].
A decade after the first GC-MS publication in 1959, Jansson demonstrated the potential
use of GC-MS for breath analysis by identifying a few breathborne VOCs, including methane,
acetone, and isoprene[68, 69]. While experiments utilizing urine headspace vapors were
analyzed to generate metabolic profiles[70-73], Pauling’s successful identification of 250
compounds from exhaled breath first hinted at the power of gas sensors for medical
volatolomics[74]. A subsequent experiment sought to link increased levels of sulfuric compounds
to liver cirrhosis[75]. Research by Gordon and O’Neill facilitated the incorporation of whole-
sample mass spectral matching algorithms to a public database to successfully distinguish lung
cancer patients from healthy controls based solely on expired breath[76, 77]. Further work
utilizing breath-based sensors as disease diagnostics by Phillips in the 1990s supported the notion
that altered breath profiles were indicative of cancer[78] and also revealed a correlation with
schizophrenia[79, 80]. The positive results garnered from these early studies spurred the
10
development of novel sampling technologies[81-88] and protocols[7, 33, 82, 89-91], as well as a
plethora of research into the VOCs involved in various pathogenic conditions[63, 92-115], with
particular emphasis on lung cancer[19, 87, 88, 116-133]. Experimental findings by Barash et al.,
suggest that breath-based GC-MS analysis is even powerful enough to differentiate between
disease subtypes caused by genetic mutations[134].
The resolution of GC-MS and related technologies has improved tremendously since their
inceptions, with current day systems able to extract over 500 biomarkers from individual breath
samples[36, 78, 134]. Owing to their high information content, GC-MS and other mass
spectrometry techniques have been integral research tools for identifying novel compounds
contained within heterogeneous samples. The elucidation of key compounds indicative of a
particular stimuli or associated with a disease has enabled the fabrication of highly efficient
application-specific targeted sensors. Moreover, VOCs common to both in vivo and in vitro
experiments have supported the possibility of utilizing cell culture headspace as a proxy for
exhaled breath under specific environmental conditions[135, 136]. This finding has facilitated
proof-of-concept research on a broad range of novel gas-sensing devices for pathogenic and
nonpathogenic VOC-profile analysis without the need for animal or human subjects[137-156].
GC-MS has shown comparable performance to standard diagnostic imaging techniques, such as
CT and PET scans, while avoiding patient exposure to doses of ionizing radiation[157-161] and
potentially reducing the associated high rates of false positive diagnoses[162, 163].
However, GC-MS remains largely sequestered to research laboratories owing to a host of
undesirable attributes for a clinical diagnostic. The operation of such instruments especially for
highly complex samples, such as breath, requires expert personnel. For any given sample, the
11
technician must consider a myriad variables such as the operation mode (split vs. splitless),
column oven temperature, type and velocity of carrier gas, thickness of the stationary phase,
column length and diameter, etc.[65]. Additionally, the resolution and signal-to-noise ratio are
positively correlated with the length of the capillary column. This necessitates longer sample
processing times for heterogeneous samples in order to attain minimally separable
chromatographic peaks[19]. Though GC-MS is touted as a highly sensitive analytical technology,
sample pre-concentration is necessary for breath-based VOCs. This is due, in part, to the low
concentration of biomarkers in the breath as well as the inefficiency of the mass spectrometer’s
electron ionization process, which only ionizes 0.01-0.001% of the total analyte molecules[65].
Furthermore, the high oven temperatures restricts GC-MS to analyzing compounds that can
sustain thermal vaporization without chemical decomposition[65, 164]. While GC-MS is not likely
to be of particular use as a point-of-care clinical diagnostic, it has conclusively demonstrated that
breath-based biomarkers, potentially indicative of an underlying pathophysiological condition,
are well within detectable ranges. As related instrument technology continues to mature, greater
degrees of sensitivity will be possible, unveiling novel chemical species that may aid in our
understanding of the underpinnings of a variety of disease states[65].
Ion Mobility Spectrometry (IMS) & High-Field Asymmetric Waveform Ion Mobility
Spectrometry (FAIMS)
Another type of untargeted gas-sensing technology, ion mobility spectrometry (IMS), has also
proven useful for disease detection and differentiation by using volatiles from exhaled breath
samples (Figure 1.6)[165-167]. As the name suggests, this technology initially ionizes sample
molecules, which allows for analyte separability based on the different mobility constants unique
12
a)
b) c)
Figure 1.6 | Ion mobility spectrometry (IMS). a) Schematic illustrated IMS principles. Molecules
are ionized in the ionization region before being intermittently passed to the drift tube. A carrier
gas is applied in the opposite direction of ion flow, whose molecules collide with sample ions at
differential rates based on cross-sectional area. Reproduced from [168]. b) Image of IMS drift
region housing drift tube attached to reaction region housing faraday plate for ion detection is
also depicted. c) Internal drift tube shown with attached electrical resistors to augment injected
current in ring electrodes. Reproduced from [169].
to each type of ion. Most commonly, these ions pass into a trapping region where packets of
them are intermittently released into the drift tube (Figure 1.6a). Within the drift tube, ions are
exposed to a homogeneous electric field, permitting a higher drift velocity in ions with higher
charges. Simultaneously an inert drift gas is propelled in the opposite direction of the electric
field and ion trajectories. Larger ions tend to engage in more collisions with neutral drift gas
molecules lowering their velocity relative to smaller ions. Therefore, once the ion gate is opened,
the ion’s overall mass and charge will determine the overall rate of migration and arrival time at
the end of the drift tube. Here, spatiotemporally stratified ions can be detected to form a mobility
spectrum or undergo additional analysis via the incorporation of tools like mass spectrometers.
IMS has two important similarities to gas chromatography: the device can be operated at
atmospheric pressures and the drift tube length determines spectral resolution. This latter
feature constrains the miniaturization of drift tube-based IMS systems, especially for complex
sample analysis (Figure 1.6b). While systems with circular drift tubes can be constructed to
mitigate the resolution-size tradeoff[170], other variants of IMS, such as high-field asymmetric
13
waveform ion mobility spectrometry (FAIMS), are more amenable to device miniaturization
(Figure 1.7) [168]. In FAIMS, molecules of a carrier gas are ionized to form reactant ions, which
collide with and bind to analyte molecules. These cluster ions are pulled into the filter region by
a) b)
c)
d)
Figure 1.7 | High field asymmetric waveform mobility spectrometry (FAIMS). FAIMS technology
is related to IMS but removes the need for an elongated drift tube and shutter, promoting device
miniaturization and continuous stream injection. a) Schematic of FAIMS-based operation inside
Owlstone Nanotech’s µFAIMS analyzer. Molecules are ionized in the ionization region (not
shown) and are transmitted through the FAIMS filter. The ions that follow stable flight
trajectories and reach the detector electrode are determined by the radio frequency input as
well as the compensation voltage. Ionization region and detector electrode are separated by a
300 µm gap. Individual electrode columns are 30 µm in width. Reproduced from [171]. b) Image
of Owlstone Nanotech’s µFAIMS analyzer. Reproduced from [171]. c) FAIMS working principle
relies on the differential mobility of various ion species. Those with similar mobilities will pass to
the detector electrode at the same compensation voltage. A full sample mixture can be analyzed
by sweeping through a range of compensation voltages. Reproduced from [172]. d) Owlstone
Nanotech’s Lonestar with Atlas sampling system incorporating the µFAIMS analyzer for sample
testing. Reproduced from [173].
14
the electrical field between the top and bottom electrode[174]. FAIMS-based sensors
incorporate strong oscillatory radio frequency fields which alters the flight trajectory of the
cluster ions. Simultaneously, a low-strength direct current compensation voltage is applied
between the top and bottom electrode. By changing the compensation voltage, different types
of ions will follow sufficiently stable flight trajectories to traverse the filter region and measured
at the detector. Ions that follow unstable flight trajectories collide with and are neutralized by
one of the electrodes. Sweeping this value through a preselected or experimentally determined
range can produce a plot of ion current intensity as a function of the compensation voltage.
IMS-based devices are relatively new in the field of diagnostic volatolomics. Traditional
applications of IMS were predominantly concerned with drug detection, explosive detection, and
environmental monitoring[168]. One of the earliest studies applying an IMS sensor as a
diagnostic tool demonstrated significant differences between the chromatograms of healthy
individuals and those with lung infections[175]. Subsequently a number of studies have sought
to provide support for IMS as a clinical diagnostics for a range of pneumological and nephrological
conditions as well as those common in critical care[176]. Mobility spectrum chromatograms from
healthy individuals were shown to be differentiable from those of individuals with lung
cancer[177-180] and other diseases[166, 181-187]. Some evidence suggests that IMS devices can
also be used to differentiate between disease subtypes due to genetic mutations[188]. As of late,
FAIMS devices have been investigated owing to their attraction as point-of-care devices. They
have proven effective in identifying various cancers from breath samples and urine
headspace[97, 173, 189-191]. A number of commercial products have been developed, some of
which are currently being tested in clinical trials (Owlstone, Thermo Fischer, Chemring, Bahia 21).
15
Having been routinely utilized for field-based applications, IMS and FAIMS sensors are
quite robust and operable in a variety of environmental conditions. Additionally, the processing
time for IMS and FAIMS devices is on the order of seconds to minutes, offering online or quasi-
online sample analysis[168, 171, 192-196]. GC-MS requires sample pre-concentration on
trapping materials before offline analysis. In contrast, IMS devices avoid this requirement as their
detection limits can progress into the parts-per-trillion ranges[168]. Hyphenated IMS techniques
with in-line pre-concentrators can even offer sensitivities in the parts-per-quintillion range[197].
A common struggle for breath-based gas sensing technologies is the high levels of moisture
present in exhaled breath, capable of decreasing device sensitivity by more than 40%[198-200].
The incorporation of multi-capillary columns increases water retention time, enabling a gas
chromatographic-like form of molecular pre-separation[175, 195, 198]. However, the removal of
humidity from a breath sample can wash out many organic compounds and shift peak positions,
significantly decreasing analyte identification accuracy[176, 201].
IMS and FAIMS technologies are powerful tools for analyzing gaseous stimuli, yet they are
not without their drawbacks. These systems are restricted to only ionizable compounds and,
unlike GC-MS, there is no general database against which to compare mobility spectrum
chromatograms. Sample ionization often introduces radiation hazards[168, 197] and readouts
for such technologies are rather complex, requiring skilled operational personnel. However, the
largest concern when employing IMS and FAIMS devices is their limited resolution. Ions can
cluster and interact in the ionization region (and drift tube if present) leading to overlapping
chromatographic peaks[172, 197]. High resolution is likely critical for highly complex breath
samples that can contain well over 500 compounds while only minute differences in a fraction
16
may be relevant for diagnostic purposes. Novel ultra-high-resolution ion mobility
spectrometers[202] and the selective use of hyphenated techniques can aid in analyzing specific
chemicals present in complex mixtures, yet whether this will lead to a preference for IMS over
GC-MS remains to be seen[172].
Electronic Noses
Untargeted chemical sensors such as GC-MS and hyphenated IMS/FAIMS can provide excellent
insight into the underlying biochemical structure of a gaseous sample. Stand-alone IMS/FAIMS,
while sacrificing specific analyte identification, permits portability, lower power requirements
and a considerable degree of separation strength. Such technologies are sensitive enough to
detect a large number of trace level chemicals present in exhaled breath, sometimes even
without sample preconcentration. However, for complex diseases involving subtle concentration
changes of numerous VOCs, these component-wise methods may not offer sufficient whole
sample discriminability. Electronic noses, or enoses, strive to generate a qualitative readout of
sample gaseous stimuli in its totality. Originally inspired by biological olfaction, they seek to
incorporate numerous cross-reactive sensors and sophisticated signal compression algorithms,
thereby generating stimulus-specific fingerprints. These “breathprints” can be conceptualized as
a point in high-dimensional space, in which the number of dimensions is a function of the number
of sensors[203]. The system design and combinatorial selectivity eliminates the necessity for high
sensor specificity and engenders an immense encoding hyperspace, capable of classifying highly
heterogeneous stimuli[204].
The introduction of cross-reactive sensors reduces the impact of confounding factors,
such as inter-individual variability in breath profiles[205], since alterations of one or a few VOCs
17
will have a negligible effect on the overall compound signal[36, 102, 206, 207]. Importantly,
enose sensitivity and specificity correspond to the total amount of nonredundant information
procurable by each individual sensor. As such, simply incorporating a higher sensor concentration
density without diversifying response characteristics serves to increase overall system noise[208,
209]. This critical need for targeted application specificity in sensor design in conjunction with
significant improvements in nanotechnology, microfabrication and materials sciences have
fueled the development of novel enose systems.
The first enose, constructed by Persaud and Dodd in 1982, consisted of two tin oxide
semiconductors that elicited a conductance change once analyte molecules interacted with
ambient oxygen atoms on the metallic surface[210]. This metal-oxide semiconductor-based
sensor encouraged future metal-oxide enose iterations to incorporate a variety of doped metals
to alter analyte specificity. In these devices, combustion reactions occurring on the surface of the
metal oxide particles, oxidize or reduce the host material and issue a change of electrical
resistance (Figure 1.8d) [211]. By operating these enose devices at temperatures ranging from
200-500°C, reaction rates are increased, thereby enhancing sensitivity and selectivity[211, 212]
assuming VOC breakdown is avoided[207]. Metal-oxide sensors have shown promising results for
differentiating breath and urine samples of cancer patients and healthy controls[213-227].
Additional studies demonstrated metal-oxide sensors were effective in identifying conditions
such as diabetes[228, 229] and renal failure[228, 230]. Yet such systems suffer from specificity
limitations and are sensitive to environmental conditions[208, 211, 231].
18
a) b) c) d) e)
Figure 1.8 | Design schematics for various electronic nose technologies. a) Chemiresistor based
on monolayer-capped metal nanoparticles. b) Chemiresistor based on single-wall carbon
nanotubes c) Chemiresistor based on conducting polymers. d) Chemiresistor or chemicapacitor
based on metal oxide film e) Quartz microbalance with selective coating. Reproduced from [27].
Devices incorporating conductive organic polymers offer an alternative to metal-oxide
based enoses (Figure 1.8c). Here, the conducting polymers serve as the analyte sensing and signal
transduction element, with a pair of electrodes and a substrate material providing signal
transmission and structural stability, respectively[232]. These systems pass a voltage across their
electrode pair, eliciting current flow in the attached conducting polymer[233]. Adsorption of gas
molecules to the polymer surface alters the electron flow, imparting a measurable change of
electrical resistance[234]. As a medical diagnostic, conductive polymer-based enoses have been
used to differentiate cancer patients from healthy controls[118, 235-241], though it is important
to consider patient history prior to sampling as confounding factors could potentially skew breath
profiles[242-244]. Such devices have also been investigated as a diagnostic for diabetes[245],
asthma, and other conditions[246-248]. Enoses based on conductive polymers have been shown
to be highly susceptible to water vapor, thus complicating online, point-of-care sampling[234].
Another common enose design is based on quartz-microbalance technology, which
incorporates organometallic compounds for chemical adsorption (Figure 1.8e). Upon adsorption,
the gas molecules alter the mass and therefore the fundamental oscillating frequency of a quartz
crystal. Chemical modifications to the associated organometallic material can adjust the
19
selectivity of individual microbalance sensors. Such enose devices were among the first to
demonstrate breath-based discriminability between cancer patients and healthy controls[249].
Subsequent studies indicated that quartz microbalances conjugated with sensitive
metalloporphyrins can differentiate between cancer samples, healthy samples, and those of
other non-cancer diseases[118, 249-260]. Interestingly, early stage classification was superior to
that of later stages, suggesting unsuitability as a stage invariant disease diagnostic[252].
More recently, sensors based on various nanomaterials, such as nanocrystals,
nanoparticles, or nanowires, have been incorporated into enose devices. The nanoscale
dimensions not only promote device miniaturization, but the increased surface-to-volume ratio
increases the rate of sensor—analyte chemical interactions[261, 262]. While individual devices
vary in design specifics, all nanomaterial-based enoses consist of organic films or functional
groups that coat the exterior of inorganic conductive nanomaterials to provide VOC adsorption
sites. However, the manner of signal transduction can provide select system designs with unique
sensing qualities, which may be advantageous in analyzing samples of a particular nature. For
example, chemiresistive sensors utilize nanoparticles (Figure 1.8a), carbon nanotubes (Figure
1.8b), semiconducting nanowires, or a combination of such nanomaterials. Whereas carbon
nanotubes and semiconducting nanowires display high sensitivity for polar compounds, they
show little affinity for nonpolar molecules. In pre-clinical testing, devices incorporating these
nanomaterials have demonstrated an ability to distinguish healthy individuals from those with
various cancers[19, 101, 102, 106, 109, 114, 116, 134, 138, 263-273] as well as non-cancer
diseases[274-277].
20
While the majority of preliminary enose volatolomics research was geared towards
identifying volatile signatures of many common microbial pathogens[273, 278-310], recent
research has focused on the utilization of enose devices as diagnostics for a variety of
noncommunicable diseases. Enoses have a number of advantages inherent to their unique
system design. Cheap costs, operation simplicity and real-time readouts offer a more ubiquitous
appeal as point-of-care devices in primary and secondary clinical care. The multitude of e-nose
designs offers significant flexibility. Yet, it may be difficult to determine the best sensor for a
given task. Prior identification of key volatiles, through technologies such as GC-MS, can guide
system design, increasing sensitivity and specificity to the molecules of interest. A single breath
sample can contain hundreds of types of analytes with wildly differing chemical attributes. As
such, a complementary approach by combining various bulk and nanomaterials may aid in
a) b)
Figure 1.9 | Cross-reactive sensors enables a highly efficient combinatorial encoding scheme.
a) Left, four sensors can encode a maximum of four stimuli in a labeled-line approach. Right, the
same four sensors, working as a group and implementing different activation (ON-OFF)
combinations, can encode a maximum of 15 stimuli. Adapted from [311]. b) Cross-reactivity of
sensors enables sophisticated pattern recognition algorithms to map stimuli to unique parts of a
high dimensional encoding hyperspace. Dimensionality reduction techniques can be utilized to
reduce hyperspace dimensionality and enable response visualizations in an algorithm-specific
subspace. Adapted from [1].
21
improved feature extraction and sample identification[312]. E-noses replace the intricate
molecular separation techniques seen in more conventional technologies with cross-reactive
sensors (Figure 1.9), promoting device miniaturization and portability. The inclusion of sensors
capable of responding to numerous analytes also endows e-noses with a robustness to
environmental variability and background interferants. This feature is critical for applications
such as breath analysis, in which the concentrations of volatile metabolites can be highly
codependent. Enoses target the relative abundance of various biomarkers, which may be much
more indicative of an underlying pathology than exact concentration changes.
The low-profile nature of enoses makes them an attractive gas-based diagnostic but
enforces limitations. For the ideal point-of-care medical diagnostic, no pre-processing steps, such
as the concentration of volatiles onto a sorbent material, will be necessary. Sensitivity is of
particular concern when analyzing trace-level VOCs present in breath samples with any
technological intervention. The majority of enose devices exhibit sensitivities in the parts per
million and upper parts per billion ranges[313]. High relative humidity is a challenge for many
technologies and enoses are no exception. Varying degrees of water vapor in each breath sample
complicates system calibration and can induce variation in sensor responsivity[204, 209, 312].
The struggle to achieve high sensitivity as well as process humid samples can lead to significant
losses of potentially useful chemical information. Moreover, sensor drift, characterized by a
gradual change in output despite no change in input, can corrupt signals and accurate stimulus
classification[231, 314, 315].
When designing an effective gaseous chemical sensor, it is essential to integrate
technologies from discrete scientific domains. In order to optimize performance, the component
22
limiting device efficacy must be identified and improved[316]. The sensor material interacts with
the volatile molecules and, thus, serves as the first and most significant bottleneck for stimulus-
related information. Moreover, the sensitive materials seem to exhibit an inherent tradeoff: high
levels of specificity correspond to high levels of irreversibility[312]. Enoses attempt to find an
optimal balance between the two so as to enhance device stability and longevity. However, this
approach seems to significantly limit their chemical resolvability. Biological olfaction, from which
enoses are inspired, has seemed to favor the latter, given the fact that human receptor cells only
have a lifetime of a few weeks[317]. Olfactory receptors have also been shown to be more
broadly tuned than the synthetic sensors of enoses, relying to a greater degree on their
population-wide activity for odorant processing[318]. Apart from encoding stimuli as a single
entity, these critical distinctions suggest that enoses bear little resemblance to biological
olfaction[211]. Considering that living organisms are the ultimate sensing machines, it reasons
that the demanding conditions of stimulus discrimination in the natural world have imposed
stringent constraints on effective chemical sensors[319] that human manufacturing has yet to
match.
Biological Olfaction
Olfactory-based chemical sensing is an inordinately difficult task, mandating precise detection
and efficient encoding of millions of different odorants with strikingly disparate molecular
configurations[320]. Yet, it can provide critical information about the surrounding chemical
environment, including the presence and location of food, predators, and potential mates. For
example, animals must differentiate between vegetation of different genotypes and growth
stages to determine if they contain a sufficient nutritional payload. Foraging insects identify
23
flowers that have already been visited to maximize the probability of reward procurement while
minimizing energy expenditure[321-324]. More extremely, all organisms must differentiate
between toxic and innocuous stimuli for proper growth and development[319]. In fact, the
benefits of chemical sensing are so valuable that olfaction was one of the earliest senses to evolve
and striking commonalities are seen between genetically dissimilar organisms[325].
In vertebrates, hydrophobic odorant molecules initially enter the nasal epithelium where
they are bound and solubilized by odorant binding proteins. These proteins transport the
molecules to the superficially located cilia of olfactory receptor neurons (ORNs). Here, binding
affinity is dictated by the underlying chemical configuration of the odorant molecule as well as
the receptor expressed on the cilia. The number and diversity of functional odorant receptors
(ORs) permits substantial specificity to a broad range of chemicals and is thought to be indicative
of overall olfactory ability[326]. Humans, for example, possess 388 functional OR genes (802 in
total), enabling olfactory-based discrimination of up to more than one trillion stimuli[1, 327-329].
Canines, whose olfactory sensitivity is 100,000 times stronger than that of humans, display 713
functional OR genes (971 in total)[326]. Each ORN expresses only a single type of OR, however
ORs can exhibit significantly different tuning curves, from acutely specific to very broad. One
theory posits that ORs recognize particular features of odorants, with some being more
ubiquitous than others, to allow for variable chemical recognition specificity. The axonal
processes of all ORNs expressing a particular receptor innervate olfactory bulb neurons within
one of many glomeruli, producing a topographically-specific epitope map[330, 331]. Owing to a
dense network of inhibitory granule and periglomerular cells, glomeruli function as highly
efficient information processors, providing a 100-fold decrease in the number of neurons
24
necessary for signal transmission[332, 333]. This is accomplished by incorporating an oscillatory-
based temporal component to the spatially distinct glomeruli, dramatically expanding the
odorant encoding hyperspace (Figure 1.10) [334-336]. Extrinsic neurons, mitral and tufted cells,
propagate action potentials to higher order brain areas that further interpret the incoming
electrical signal[337].
The skeletal system offers immense structural stability, ultimately permitting vertebrates
unique developmental adaptations and ways in which to interact with the physical
environment[338]. Alternatively, insects utilize an exoskeleton, limiting growth potential and
Figure 1.10 | Temporal encoding rapidly expands odor state space. The use of multiple sensors
enhances the encoding state space in an exponential manner. Adding a dynamic temporal
component, increases the encoding space even further. For example, using a set of only four
sensors, each with two possible states (ON-OFF), and 10 time points, allows for over a trillion
possible spatiotemporal signals (24 = 16, 1610-1 = 1,099,511,627,775).
25
total lifespan. As a result, insects have developed highly acute chemosensory abilities due to their
reliance on olfaction for nearly all tasks, from food source localization to predator and prey
detection to identifying potential mates. While many fundamental chemosensory mechanisms
remain consistent with those of vertebrates, insects have evolved a number of adaptations that
are well-suited to their ecological niches[339-341]. Odor perception in insects originates in
microscopic hairlike structures, known as sensilla, that coat the surface of the antennae (Figure
1.11c). Similar to the nasal epithelium in vertebrates, these sensilla house ORN dendrites that
interact directly with odorant molecules. Once an odorant molecule is encountered, it enters a
a) c)
b)
d) e)
Figure 1.11 | Sensilla coat the locust antennal surface. a) Antenna depicting individual antennal
segments. b) Antennal segments housing microscopic sensilla. c) Electron micrograph of distal
antennal segment depicting various types of sensilla: ch- chaetica, bs- basiconica, co-
coeloconica, tr- trichodea. d) Individual basiconic sensillum. Tiny pores along the cuticular surface
allow molecule entry and subsequent ORN-analyte binding. e) Individual trichoid sensillum. Note
the slender morphology of the trichoid sensillum vs. the basiconic sensillum in d. Reproduced
from [342].
26
sensillum through tiny cuticular pores or spoke channels, where it is solubilized and transported
to ORNs by odorant binding proteins. ORNs are the site for chemical reception and signal
transduction and, as such, are of particular importance for chemical sensing. Structural studies
Figure 1.12 | Olfactory receptors display significant variability in chemical specificity. Functional
responses of twenty-four olfactory receptors to a panel of 110 different odorants is shown for
Drosophila melanogaster. Each panel indicates tuning properties of an individual olfactory
receptor. The tuning curves are ordered according to the number of odorants eliciting a strong
response. Through combinatorial coding, variable tuning properties allow for discrimination of a
large number of diverse odorants at low concentrations. Reproduced from [343].
27
have demonstrated that densities of olfactory receptor sites can reach 30,000 per µm2[344]. ORN
response properties are dictated by the expressed olfactory receptor and can exhibit broad
molecular specificity, responding to hundreds of different analytes, or very high selectivity with
strong binding affinities for one or a select few (Figure 1.12). Insects display a number of
morphologically and functionally disparate olfactory sensilla. Sensilla basiconica, the most
abundant type, contain a high density of pores, whereas sensilla trichodea contain far fewer,
thereby limiting the rate at which molecules can enter and bind to ORNs. Moreover, the internal
configuration of sensilla appears to be species-specific. Whereas sensilla basiconica house either
two or four ORNs in Drosophila melanogaster[345], in locusts they can house anywhere between
20 and 50[342]. The presence of more neurons within an individual sensillum introduces a
competitive ligand binding environment subsequently altering the ionic concentration gradient
experienced by any one neuron. Unlike the metabotropic OR receptors found in vertebrates,
evidence suggests that invertebrate ORs form heteromeric ionotropic transmembrane channels
that allow direct ion influx upon ligand binding[346, 347]. This theory has been corroborated by
molecular studies[347, 348] as well as the low-latency of odorant-based electrophysiological and
behavioral responses observed in a variety of insect species[349-351]. ORNs exhibit low levels of
spontaneous activity, but upon ligand binding, response dynamics of individual neurons and
entire neural ensembles shift dramatically. The spontaneous baseline activity presents an
ingenious method for increasing the chemical encoding state space. Not only can odorant binding
cause unique upregulations in spiking activity, but others, oftentimes aversive odorants, can
significantly reduce firing rates. While at the cost of a slight increase in energetic output
necessary for the spontaneous activity of these neurons, an individual neuron can demonstrate
28
additional unique firing patterns to incoming stimuli. Efferent signals are transmitted along the
ORN axons into glomeruli of the antennal lobe[352]. Most insects, including Drosophila
melanogaster and Apis mellifera, project receptor-specific ORN axons to one or two glomeruli
within the antennal lobe, closely resembling the uniglomerular branching patterns observed in
the olfactory bulb of vertebrates[353-358]. However, the antennal lobe of most insects contains
an order of magnitude fewer neurons and glomeruli than are found the vertebrate olfactory bulb.
Here, a complex network of inhibitory local neurons induces 20-30 Hz oscillatory cycles that are
essential for odor recognition[359, 360]. Additionally, the signal processing within the antennal
lobe has been shown to amplify weaker signals as well as mitigate undesirable chemical
noise[361, 362]. This oscillatory behavior not only synchronizes the activity of efferent projection
neurons to transmit action potentials to higher-order processing regions, but also increases the
total number of encodable stimuli. Inclusion of a temporal component seems to be essential for
providing organisms with high olfactory resolution, necessary for interpreting complex stimuli
and differentiating similar odorants[363, 364]. Afferent signals become increasingly sparse in
higher order brain regions as neuronal divergence occurs and neuronal tuning curves show very
high signal specificity[360, 365-367].
While the low number of neurons in the insect olfactory system seemingly ought to endue
gross detection limits and poor chemical differentiation relative to vertebrates, insects are
equally or more potent in their sensitivity and discrimination capabilities[368]. Separate studies
have found moths to elicit an observable physiological response to a single or tens of chemical
molecules[369, 370]. Exactly how such low detection limits are achieved is an active area of
29
Figure 1.13 | Olfactory receptors display significant variability in chemical specificity. The red
and blue bars represent the number of active and pseudogenes, respectively. Vertebrates exhibit
significantly more olfactory receptors (OR) than insects. Moreover, they have a larger portion of
disrupted pseudogenes that no longer encode for functional olfactory receptors. This suggests
the development of and heavy reliance on sensory abilities other than olfaction. While insects
display approximately an order of magnitude fewer olfactory receptors than vertebrates, they
exhibit alternate coding schemes enabling highly acute chemosensation. Reproduced from [371].
research. The heavy reliance on olfaction has caused insects to retain many more functionally
active ORs than their vertebrate counterparts (Figure 1.13). The limitation of expressing a low
number of neurons for sensory processing is likely also significantly mitigated by the intricate
network dynamics observed in the antennal lobe. One key anatomical difference is that ORN
axons within the olfactory bulb primarily innervate extrinsic neurons directly as well as indirectly
via interneurons, such as granule cells[372]. In the antennal lobe, however, ORNs only form
polysynaptic connections to projection neurons by way of local neurons, with no direct
input[368]. Additionally, extraglomeruluar synapses are present in the olfactory bulb, whereas
none have been observed in the antennal lobe[333, 368]. Exactly how these anatomical
30
disparities alter functional aspects of olfactory encoding remains to be seen, yet insects have
been found to display exceptionally high levels of olfactory acuity based on a significantly reduced
neural network.
Biosensors & Bioelectronic Noses
The prowess of animal olfaction has been long recognized by humans. The domestication of dogs,
for example, first allowed humans to take advantage of the canine olfactory system and
behavioral responses for prey localization during hunting or fumes indicative of dangerous
environments. The sensitivity, selectivity, generalizability and response rapidity of canine
olfaction cannot be matched by current state-of-the-art manmade sensors[233].
Understandably, canines have proven highly effective for a number of applications from the
detection of explosive compounds[373-378] to identification of human disease states[379-386].
Since the first anecdotal reports of canines’ abilities to recognize various forms of cancers in their
owners[387-390], research utilizing sniffer dogs as medical diagnostics has shown promising
findings. Though results vary due to lack of standard training and testing regimens, canines have
demonstrated impressive discriminatory power in identifying samples associated with lung,
breast, and skin cancer, among others[225, 391-406].
In addition to expressing more ORs than humans, dogs’ enhanced olfactory abilities stem
from a substantial increase in nasal epithelium surface area, permitting a higher density and 20-
40 fold more total ORNs (Figure 1.14) [355, 407]. As a result, the sensitivity and specificity of
canine olfaction, especially among variable background interferants, is exceptional, with
detection levels at 1-2 parts-per-trillion[391, 399, 408-410]. Thus, it is understandable that the
31
Figure 1.14 | The highly convoluted nasal epithelium of canines enables impressive olfactory
abilities. The structural adaptation of a enables a significant increase in ORN density. In
conjunction with the wide variety of ORNs, canines can detect a broad range of chemical stimuli
with sensitivities as low as 1-2 parts-per-trillion. Reproduced from [411].
sensitivity of artificial systems pales in comparison to biological olfaction, given that most devices
contain a maximum of a few dozen sensors. Furthermore, the high water vapor concentration
present within biological samples, while a significant concern for analytical devices, actually
enhances canine olfactory abilities[412]. Nevertheless, sniffer dogs for disease detection are not
without their limitations. Since canine chemical reception is inherently linked with respiration,
only approximately 12-13% of inspired air reaches the olfactory neuroepithelium, thus wasting a
fair amount of otherwise useful chemical information[411]. Highly standardized training
protocols that often last weeks or months must be adopted for a particular stimulus[413, 414].
Furthermore, easily interpretable binary behavioral responses limit readout intricacies and the
amount of stimulus-related information. Other models utilizing rodents to detect tuberculosis via
32
sputum samples have elicited positive results with minimal testing time relative to canines[415-
419]. While rodents express even more ORs than canines and are not easily influenced by a
specific handler, they face similar sampling and readout challenges. Animal models mitigate the
sensing limitations seen in manmade technologies, however reliance on behavioral readouts
significantly hinders signal saliency.
To bypass many of the limitations of using current animal models as chemical sensors,
researchers have looked to invertebrates, which exhibit a much wider dynamic range than that
of their vertebrate counterparts[420]. A few studies have determined that the nematode,
Caenorhabditis elegans, can be genetically modified to reliably exhibit stereotypical chemotactic
behaviors when presented with explosives[421], prominent tuberculosis biomarkers[422], and
cancer urine samples[423-428]. The parasitic wasp, Microplitis croceipes, has been explored for
its use as a gaseous sensor for a variety of VOCs[429-431]. Detection limits were found to be 10-
fold lower than those for a common electronic nose[432]. While displaying very high sensitivity,
the wasps did not respond to some standard alcohols[429, 433], suggesting a limited range of
chemical specificity.
The chemical sensing abilities of higher order insect species exhibiting more complex
olfactory circuitry and discriminatory power have also been investigated. Field studies with
honeybees, Apis mellifera, have shown them capable of detecting explosive vapors at parts-per-
billion and parts-per-trillion concentrations, surpassing state-of-the-art devices based on
IMS/FAIMS technologies[434-437]. For medical purposes, sniffer bees can be taught to elicit a
proboscis-extension behavioral response when presented with various compounds associated
with tuberculosis[438]. A prototype breath sampling device has been developed intended to
33
screen patients for various diseases based on the behavioral responses of pre-conditioned
honeybees[439, 440]. Like canines, bees have evolved in naturalistic conditions, permitting
excellent detection sensitivity for humid samples or samples among humid environments[434].
In the moth, Manuduca sexta, King et al. showed that olfactory-induced electromyographic
signals could be used as an electrical proxy for a behavioral readout[441]. The technological
application allowed for a finer level of discrimination than binary behavioral responses, however,
the insect still necessitated a training period to associate muscular activation with the processed
neuronal signals. While insects do not require significant training times (hours-days)[442], basing
stimulus identity on behavioral readouts, whether vertebrate or invertebrate, requires that they
be preconditioned to associate a reward with the intended stimulus.
In an effort to remove this bottleneck, biohybrid chemical sensors have predominantly
sought to incorporate low-level neural correlates, such as specific olfactory receptors and even
entire insect antennae. Antennal-based electronic noses target either neural activity from a large
number of ORN axonal processes across the antennal flagellum or a more select group of ORN
dendrites located within individual sensilla. For whole-antennae electroantennography (EAG),
the electrical potentials of ORN axons are recorded by metal or saline-filled glass electrodes.
Despite reflecting the aggregate activity of a large neural population, resulting readout
amplitudes and shapes roughly correspond to the concentration and identity of the presented
odorant stimulus[443-445]. A number of EAG technologies have been developed since its
inception, including portable antenna-on-a-chip devices[444, 446-448], multi-antennae
arrays[449-452], and hyphenated GC-EAG systems[449, 453-456]. EAGs have largely been used
to investigate insect response profiles to a variety of pheromone components, given their strong
34
attunement to pheromone blends[457]. Other EAG devices have also shown promise for more
general chemical-sensing applications[458-463], detecting parts-per-trillion concentrations for
some odorant species with the incorporation of signal amplification technologies[464-468]. In
addition to chemosensation, various types of sensilla respond to a variety of input modalities,
such as temperature, humidity, and mechanical stimulation. The non-specific recording
technique employed by EAGs necessitates stringent environmental controls to mitigate the
influence of such non-chemical signals. The longevity of such devices has also been of concern.
Whereas antennae-only setups enable device miniaturization, satisfactory readings last for
approximately 30 minutes[446]. Whole animal EAG recordings, in contrast, can be continued for
several days and are ultimately limited by the total lifetime of the incorporated organism[443].
The main limitation of electroantennogram technology stems from the fact that the electrical
potentials represent the summation of activity of a large population of ORNs. This reduces the
overall chemical specificity as individual ORN responses, which may contain feature-specific
information, likely only minorly contribute to the aggregate population-wide signal. As such,
information from sparsely firing or smaller populations of narrowly tuned neurons that would
otherwise transmit critical stimulus features to the antennal lobe may be irresolvable[449].
Alternatively, sharp metal or glass electrodes can be inserted into individual sensilla to
record electrochemical alterations within the extracellular lymphatic solution surrounding ORN
dendrites to enhance chemical specificity. These sensilla-specific recordings (SSRs) offer lower
detection thresholds and effective neuron differentiation via spike amplitudes[449, 455, 469-
473]. Recordings from diverse odor panels have demonstrated rich temporal response dynamics
based on stimulus identity, with each neuron responding to a limited subset of odors[457, 474-
35
482]. SSRs even elicit clear responses to a number of natural stimuli as well as biologically
irrelevant odorants, such as those from explosives and illicit drugs[483, 484].
Whereas EAG recordings are characterized by their gross neural population-wide
resolution, SSRs occupy the alternate end of the spectrum. The high spatiotemporal precision
enables intricate analyses of fine-scale biochemical processes occurring within individual sensilla;
however, it also significantly restricts scaling up associated recording systems to attain a broad
response profile from disparate sensilla. Moreover, the small profiles and high impedances of
SSR electrodes make them vulnerable to signal contamination via electrical interference.
In addition to electrode-based recordings, transgenic organisms and fluorescent dyes can
be used to visualize intracellular calcium concentrations that are indicative of underlying neural
activity. In vivo calcium imaging of the antennae of fruit flies bypasses the challenge of
simultaneously targeting multiple disparate sensilla by incorporating a relatively wide field of
view. This sensillar-ensemble technique has been used to successfully discriminate between
various odorants[485-487] as well as the volatiles associated with cancerous and healthy cell
cultures[488]. However, most imaging techniques for odorant encoding in insects have targeted
second-order neurons within the antennal lobe. Projection neurons, the efferent cells of the
antennal lobe, exhibit a number of key features elevating their ability for stimulus encoding and
candidacy for bioelectronic interfacing. Relative to ORNs, PN responses show superior reliability
and are more broadly tuned, increasing the stimulus related information attainable from any one
cell over time[489-491]. This is in large part due to the intricate inhibitory network of local
neurons as well as reciprocal PN-LN interactions, which are responsible for sublinear signal
transformations between ORNs and PNs[491-496]. Calcium imaging of antennal lobe PNs has
36
shown odor-specific glomerular activation in multiple invertebrate species, including
locusts[497], honeybees[498-503], and others[485, 504-507]. However, the reliance on optical
methods restricts neuronal analysis to visible glomeruli and the temporal resolution of such
technologies is too low to reveal fast dynamic effects[499-501]. While calcium imaging permits
the differentiation of various stimuli, the lack of high temporal resolution likely impairs
discriminating between complex, similar odorant mixtures. Since 20-30 Hz oscillations are
essential for efficient signal transmittal by projection neurons, technologies with courser
sampling rates will promote information loss. Calcium signals are not always correlated with
neuronal action potentials and increases in fluorescence can be observed due to subthreshold
activity[497]. Alternatively, electrode-based recordings are not restricted by the relatively slow
kinetics of calcium indicators or limiting optical parameters, such as tissue-induced light
diffraction[508-510]. Numerous studies have shown that odorant-induced projection neuron
activity can be used to discriminate between a plethora of natural stimuli using electrode-based
technology[360, 362, 491, 511-522]. Taking this approach a step further, Saha et al.
demonstrated the efficacy of an insect biosensor for identifying varying concentrations of
explosive compounds, which have played no part in guiding olfactory sensory evolution[523].
Forward Engineering Novel Insect Biosensors
Our forward-engineering approach for chemical gas sensing combines a highly sensitive sensor
with broad-ranging specificity and an efficient functional processing scheme optimized to elicit
highly discriminable spatiotemporal response patterns. The decoupling of olfactory receptors
and/or antennae from the dense network of the antennal lobe would not be a significant
bottleneck if the inner dynamics of biological olfactory systems had been sufficiently elucidated.
37
Even more extreme, researchers pursuing gas-sensing devices based on cross-reactive sensor
materials, exclude biological sensors and, therefore, any downstream circuitry. This has
contributed to the inability of enoses, for example, to process natural stimuli as rapidly and
efficiently as the biological olfactory system it aims to emulate. By tapping into the extrinsic
neural signals of the antennal lobe, we posit that we can interpret the intricate spatiotemporal
response patterns as encoding for key volatile biomarkers seen in the exhaled breath of diseased
patients. Moreover, we suspect that the massive encoding state space affiliated with such
processing schemes is powerful enough to differentiate between small, few-component
variations in highly heterogeneous mixtures. Our rationale for intervening at the point of the
antennal lobe is rooted in functional neurobiology. The substantial convergence of chemical
signals to a relatively small number of extrinsic neurons permits maximal information extraction
by a finite number of microelectrodes. The spatiotemporal response patterns are amenable to
high-level, time-based signal processing analyses and ultimately may underly a gas sensing
system with discriminatory power superior to that of current state-of-the-art manmade
technologies.
38
CHAPTER 2 | METHODOLOGY
In order to record precisely timed stimulus-evoked neuronal signals, a number of independent
technologies were integrated.
Odorant Delivery
A cylinder of purified zero-contaminant air provided all supply air to a commercial olfactometer
(Aurora Scientific, 220A). The olfactometer can deliver highly controlled concentrations of
gaseous odorant stimuli with precise timing. Upon sequence initiation, a stream of clean supply
air equal to that of the experimenter-determined total air flow rate flowed through mass flow
controller three (fresh air line) via a 1/16” diameter polytetrafluoroethylene (PTFE) tube to a final
solenoid valve. Another stream of clean air was sent through the other two mass flow controllers
via similar PTFE tubing. Mass flow controller two (odor line) passed a percentage of the total air
flow rate predetermined by the experimenter. The remaining air flow equal to that of the total
air flow rate minus the odor line air flow rate was directed through mass flow controller one
(dilution line). The odor line and dilution line joined as air streams from each line mixed together
upstream of the final valve. The final valve directed the incoming air lines one of two ways: to an
exhaust line or to a thirty-centimeter-long stimulus flow line, the end of which was positioned
approximately two to three centimeters from the most distal antennal segment. During
interstimulus intervals, the vial inlet and outlet valves remained closed. Clean air from the odor
line continued through the mixing valve where it mixed with clean air from the dilution line. Air
from the fresh air flow line was directed via the final valve towards the insect antenna, while
combined dilution-odor line air was directed via the final valve towards exhaust. Five seconds
prior to stimulus delivery, vial inlet and outlet valves opened while the mixing valve closed. The
39
clean air from the odor line carried the vial headspace contents through the vial outlet valves
until combining with the clean air from the dilution line. The final valve remained in the
unactuated state, directing clean air to the insect and odor-laden air to exhaust (Figure 2.1a).
After receiving a stimulus delivery signal, the final valve was actuated, directing clean air to
exhaust and odor-laden air to the insect (Figure 2.1b). Stimulus periods lasted four seconds in
duration, at which point, the final valve returned to the unactuated state. One second after
returning to the unactuated state, the vial/flask inlet and outlet valves were closed to ensure the
a) b)
Figure 2.1 | Olfactometer flow design schematic (adapted from Aurora Scientific). Mass flow
controller 3 (MFC3) passes the total flow rate of clean air to the final valve via the clean air flow
line. An additional allotment of clean air equal to that of the total flow rate is split between mass
flow controller 1 (MFC1) and 2 (MFC2), the dilution and odor lines, respectively. The path of the
air travelling through MFC2 is dependent on the state of the olfactometer. a) Olfactometer
configuration in the unactuated state. The vial inlet and outlet valves remain closed while the
mixing valve is open. Clean air from the odor line combines with that of the dilution line to
achieve the total flow rate. The final valve directs this portion of clean air to exhaust, while
directing the clean air from MFC3 to the insect antenna. b) Olfactometer configuration in the
actuated state. The vial inlet and outlet valves open while the mixing valve remains closed. Clean
air from the odor line travels into the vial, disrupting the sample headspace, and recombines the
odor-laden air with clean air from the dilution line. The final valve directs this portion of clean air
to the insect antenna, while directing the clean air from MFC3 to exhaust.
40
headspace sufficiently repopulated with odorant volatiles prior to successive stimulus deliveries.
This protocol was designed to maintain a constant flow rate through the stimulus flow line,
thereby eliminating any potentially confounding neuronal responses due to a change in air
pressure. A six-inch diameter funnel pulling a slight vacuum was placed immediately behind the
Figure 2.2 | Odor Delivery Program. Labview (National Instruments) block diagram depicting
clock synchronization controller schematic. The three sequences depict three independent
clocks. The top row corresponds to the clock determining the absolute times to record and save
neural signals. The middle row determines the timing of the TTL pulse to begin the olfactometer
challenger sequence. The bottom row guards the triggering of the final valve on the olfactometer.
Users selected the number of trials to run. Electronic feedback loops were installed to the
accompanying DAQ assistant. The right-hand side of the block diagram displays plots for the
outgoing signals to ensure timing of each clock occurred as expected. If the total time selected
for any of the clocks was not equal, the program would not execute and throw a warning.
41
insect to ensure swift removal of odorants. Each stimulus was repeated five times with an
interstimulus interval of one minute. The order of stimuli was pseudorandomized for each
experiment.
A multifunction I/O device was installed into the computer and connected via a
multifunction cable to a terminal block. This hardware installation allowed for signal transmission
to and from external electrical devices. Custom Labview code was written (Figure 2.2) along with
a corresponding GUI (Figure 2.3). The program allowed for user-defined stimulus presentation
Figure 2.3 | Odor Delivery GUI. Labview (National Instruments) front panel depicting GUI for
selecting experimental timing parameters and initiating the sequence. Data recording times and
stimulus presentation times are customizable based on experiment. If trial times are unequal,
the program will trigger an error and will not execute. Plots on the right-hand side provide visual
feedback for signal transmission and clock synchronization.
42
periods and neuronal recording times. A high TTL output pulse served to initiate the challenger
sequence within the olfactometer software. A subsequent high TTL output pulse opened the final
valve of the olfactometer, prompting odorant delivery to the insect antenna.
Signal Recording & Electrode Preparation
A USB interface board software was used to visualize and record all neuronal signals and an RHD
USB interface board (Intan Technologies, LLC, Figure 2.4) was used for acquisition of all neural
signals. Either four or eight of the low-noise amplifier channels were active at any one time
Figure 2.4 | RHD USB Interface Board (courtesy of Intan Technologies, LLC). Circuit board
providing up to 256 amplifier channels (64 per SPI port). SPI cable extended to a small headstage
(not shown), which digitized neuronal signals recorded from microelectrodes. Digital input
channels received a high signal during the stimulus presentation period to record stimulus onset
and offset times for subsequent analysis.
43
depending on microelectrode targeting and neuronal signal quality. The sampling rate of the
board was set to 20 kHz for all experiments. Upon stimulus presentation, a TTL input pulse issued
by the olfactometer was transmitted to the RHD USB interface board to record exact stimulus
onset and offset times. The RHD USB interface board was connected to a 16-channel preamplifier
headstage (RHD2132, Intan Technologies, LLC) via a serial peripheral interface cable. A wire
adapter connected to a 16-channel DIP pin socket received neural signals via a microelectrode
array.
For all locust-specific experiments, multi-electrode arrays (MEAs) were purchased from
NeuroNexus Technologies (Figure 2.5). MEAs consisted of dual silicon shank probes with two sets
of iridium tetrodes embedded into each shank. The lower two tetrodes were located 55-101 µm
a) b)
c)
Figure 2.5 | Neuronexus Multi-Electrode Array. a) Schematic of dual shank multi-electrode array
configuration. Figure reproduced from Neuronexus Technologies. b) and c) Magnified images of
one probe used for locust neural recordings. Four individual tetrodes can be seen along the two
shanks.
44
above the tip of the shank. The upper two tetrodes were located 121-177 µm above the lower
and were not used as recordings only required minimal penetration depths. Electrodes pads had
a surface area of 11 µm2 and a center-to-center pitch of 25 µm within each tetrode.
Custom fabricated wire tetrodes were used for honeybee-specific neuronal recordings.
Ultra-fine nickel-chromium 12.7 µm diameter wire with a ~0.254 um thick polyimide coating for
insulation was used (Sandvik Technologies). A six-foot length of wire was drawn out and secured
at both ends with lab tape. The taped ends were carefully brought together while avoiding the
introduction of any kinks in the wire strands. Using tungsten-carbide scissors, the looped end
a) b)
Figure 2.6 | Tetrode Wire Drawing. a) Tungsten-carbide scissors were used to cut the wire
loop(s), creating strands of equal lengths. b) Strands were manually drawn out to ensure equal
length prior to securing with tape. Image depicts two 12.5 µm wires.
45
distal to the tape was cut, ensuring the wire strands were of equal length (Figure 2.6). Another
piece of lab tape was placed over the ends of the cut-wires and subsequently folded onto the
existing taped end. The dual-loops were strewn over a lab retort stand bar and a binder clip was
attached to the taped ends to stretch the wire strands and keep them in place overtop a magnetic
stirrer. Wires were twisted in a clockwise fashion for approximately five minutes at 500 rpm. The
twisted wire strands were then heated by running a heat gun up-and-down the length of the wire
for ten passes lasting a total of ~45 seconds (Figure 2.7). After heating, wires were allowed 20
seconds to cool, then spun in a counterclockwise fashion for five to ten seconds to release any
residual tension. Having stopped the spinning wires, another ten passes of a heat gun was applied
to the wire strands. A piece of lab tape was applied to the top of the wire near the looped ends.
Holding the tape, the loop was cut, and the wire was strewn across two elevated platforms
a) b) c)
Figure 2.7 | Tetrode Spinning. Looped wire strands were strewn over metal retort stand atop a
magnetic stirrer. Attached binder clip allowed free rotation of the lower end, spinning wires into
a tetrode. The wires were heated to reduce residual tension and prevent wire recoiling once
removed from the stand.
46
a) b) c)
d) e) f)
Figure 2.8 | Tetrode Wire Removal and Sectioning. While holding the wire between lab tape,
the loop was cut, and the twisted wire was strewn across two elevated platforms. Special care
was taken to ensure the wire was kept taught and did not exert any recoil. Both sides were
secured using lab tape.
(Figure 2.8). Tape was applied to both ends to prevent any coils or kinks from forming. Five-
minute epoxy was mixed in a small petri dish and applied at intervals along the length of the
twisted wire tetrode strands. A ball of epoxy was placed two inches (51 mm) from the future
brain insertion point of the tetrode wires. After another inch, a second ball of epoxy was
positioned to section off the portions of the wires to be soldered. The process was repeated for
the entire length of the twisted tetrode strands (Figure 2.9). The epoxy balls were left at room
temperature overnight to ensure full curing occurred. The wire strands were cut immediately
upstream of the sections to be soldered. A thin-walled PEEK tube was cut to size (Figure 2.9) and
47
a) b)
Figure 2.9 | Tetrode Epoxy Application. Equal parts of two-part epoxy were mixed, and droplets
were applied along the length of the spun tetrode. One epoxy ball was placed at the end to be
soldered and another was positioned at the junction point.
a)
b)
Figure 2.10 | Tetrode Splaying. Tetrodes were manually unwound under a stereomicroscope and
the end epoxy ball was cut off. Note: here, no junction epoxy ball is depicted to aid in tetrode
visualization. If needed, tweezers coated with a soft plastic were used to unwind individual wires
until sufficiently separated.
48
the wire bundle was placed through the tube until the end touched the remaining epoxy ball.
More epoxy was used to mechanically bond the existing epoxy ball and wire bundle to the PEEK
tube. Under a stereomicroscope, the end epoxy ball was spun counterclockwise to unwind a
portion of the wire bundle. The wire bundle was cut and any remaining wire twists were removed
by clasping the tetrode strands with a Plasti Dip coated tweezer and repeatedly running the
softened tool along the length of the wire until reaching the free end (Figure 2.10). A low-heat
electric lighter was briefly touched to the free wires to remove the polyimide insulation, while
allowing the metal wires to maintain mechanical strength and integrity. Solder flux was placed in
a) c)
b) d) e)
Figure 2.11 | Finished Tetrode. A finished tetrode is depicted having been appropriately soldered
and finished with epoxy application for structural support and additional electrical insulation. b)
Thin wires can be seen completing the electrical circuit from each wire tip to its respective pin
channel. d, e) Brain insertion point.
49
each pin connector of a four pin DIP socket prior to filling each pin connector with a small amount
of lead-free solder. Grasping each deinsulated wire strand, solder joints were remelted, and wires
were individually pulled through channel-specific solder joints until taught. Any wire extruding
beyond solder joints was trimmed. Two-part epoxy was placed over solder joints and the DIP
socket body to provide effective channel insulation and structural support (Figure 2.11).
Both NeuroNexus probes and custom fabricated wire tetrodes were electroplated to
achieve desired impedances and improve signal-to-noise ratio (Figure 2.12). A stainless-steel
anode and either multi-electrode array or tetrode were submerged into a non-cyanide gold
plating solution within a custom-designed, 3-D printed electroplating reservoir. A constant
current stimulus isolator delivered a steady state current of 1.5 mA to an IMP-2A impedance
tester, which gated the input to deliver square waves at a frequency of 3 Hz. The afferent voltage
signal was transmitted via the impedance tester to an electroswitch, which was soldered to a 16
Figure 2.12 | Electroplating setup. Current from the stimulus isolator is gated according to input
parameters on the function generator. The oscillating current is directed to the impedance
tester, which can be selected to plate electrode channels with gold ions or test the impedance of
individual channels.
50
pin DIP socket. By adjusting the channel on the electroswitch, the current was redirected to the
appropriate pin connection and corresponding electrode channel. The stainless-steel anode was
connected to the impedance tester and completed the electrical circuit. While in test mode, the
externally derived current was directed towards ground and impedance measurements of each
electrode channel were taken at 1 kHz. Upon switching to plating mode, the current stimulus
isolator-derived current was passed through the electrode channel, into the plating solution and
through the anode. Gold particles from within the solution were deposited onto the electrode,
thereby decreasing overall channel impedance. Individual channels were plated to 200-300 kW
for locust recordings and 300-400 kW for honeybee recordings.
Stimulus Creation
Odor Vials
All stimuli used for experiments were constructed within the same week to avoid volatile leakage
or chemical breakdown. For 1% odorant experiments, 10 mL of mineral oil was placed into
airtight 20 mL vials followed by the liquid chemical stimulus. For synthetic lung cancer mixtures,
10 mL of mineral oil was placed into airtight 20 mL vials with additional steps shown below.
Micropipettes were used to achieve precise chemical concentrations in individual vials.
For those chemicals naturally found in lower concentrations within healthy and lung
cancer breath samples, serial dilution was used. For simulated healthy breath samples, final
desired chemical concentrations were as follows: 0.2 µL of decane, 0.09 µL of nonanal, 0.005 µL
of 2-methylheptane, 0.0006 µL of 2,2,4,6,6-pentamethylheptane, and 0.00001 µL of pentanal
51
Figure 2.13 | Synthetic Healthy Breath Construction. Diagrammatic steps detailing the serial
dilution protocol used to create synthetic healthy breath stimuli.
(Figure 2.13). Steps to achieve appropriate chemical concentrations within odorant mixtures
were as follows:
1. 10 µL of pentanal added to 10 mL of mineral oil to make working solution
2. 100 µL of working solution added to 10 mL of mineral oil
3. 50 µL of 2-methylheptane and 6 µL of 2,2,4,6,6-pentamethylheptane added to working
solution
4. 100 µL of working solution added to 10 mL of mineral oil
5. 9 µL of nonanal and 20 µL of decane added to working solution
6. 100 L of working solution added to 10 mL of mineral oil to make final stimulus solution
For synthetic lung cancer breath samples, final desired chemical concentrations were as
follows: 0.7 µL of nonanal, 0.4 µL of decane, 0.02 µL of 2-methylheptane, 0.002 µL of 2,2,4,6,6-
52
Figure 2.14 | Synthetic Lung Cancer Breath Construction. Diagrammatic steps detailing the serial
dilution protocol used to create synthetic lung cancer breath stimuli.
pentamethylheptane, 0.001 µL of pentanal, and 0.001 µL of hexanal (Figure 2.14). Steps to
achieve appropriate chemical concentrations within odorant mixtures were as follows:
1. 10 µL of pentanal, 10 µL of hexanal, and 20 µL of 2,2,4,6,6-pentamethylheptane added to
10 mL of mineral oil to make working solution
2. 100 µL of working solution added to 9.9 mL of mineral oil
3. 2 µL of 2-methylheptane, 70 µL of nonanal, and 40 µL of decane added to working
solution
4. 100 µL of working solution added to 9.9 mL of mineral oil to make final stimulus solution
53
All 1% and synthetic lung cancer breath samples were rapidly sealed, vortexed for a total
of 30 seconds to achieve sample homogeneity, and stored in a cool, dry location prior to
conducting experiments.
Cell Culture Flasks
Modified airtight T25 flasks were used for all in vitro cell culture experiments. Initially holes were
drilled using 19-gauge needles, one in the back right corner of the body and one in the cap. One
19-gauge needle was inserted through the flask body to serve as an inlet pin and another through
the cap to serve as an outlet pin. Two successive layers of a low-volatile, two-part epoxy was
applied to the flask and cap surface, hermetically sealing the flask—needle joints. Two 5-cm long
PTFE tubes were sealed with the low-volatile epoxy, placed over both pins and stabilized via a
small tube clamp (Figure 2.15). The incorporation of the needles allowed for a rapid and simple
a) b)
Figure 2.15 | Modified Flask Construction. Images depict fully constructed flasks with cell media
and freshly seeded cells. Two layers of two-part epoxy was used to secure the 19-gauge pins to
the polystyrene T25 flasks. Flasks were always capped other than during stimulus presentation
to ensure sufficient cell culture volatiles within the flask headspace. Flasks were constructed at
least 72 hours before cell seeding to allow epoxy to cure entirely.
54
connection to an airline through which headspace volatiles could be transmitted. Five stimuli
were used in the current experiment, including three cancer cell lines, HSC-3, SAS, and Ca9-22,
one non-cancer cell line, HaCaT, and one flask with solely cell media. These cell lines were
selected due to the observed metabolic differences during longitudinal cell culture analysis
(unpublished data). Cell media (Dulbecco’s modified eagle media) was kept consistent for all
stimuli in order to maintain consistency between cell line environments. A HEPES zwitterionic
organic buffering agent was added to the flasks to promote cell growth while in a hypoxic
environment. A total of one million cells for each cell line were initially seeded in the augmented
flasks. Cell progression was monitored over the course of four days and experiments were
repeated once every 24 hours. Electrophysiological experiments took place over a period of 8
weeks. Flasks were stored in an incubator set at 37° C and were only removed when in active use
during an experiment.
Husbandry
Locusts
All locusts were housed within cages in an environmentally controlled incubator. Locust cages
were adapted from commercially purchased reptile cages to include a plexiglass front screen with
a 6” diameter hole and custom designed door covering (Figure 2.16). The locust incubator
(Powers Scientific, Inc.) operated according to a 18-6-hour light-dark cycle at a daytime
temperature of 35.5°C and nighttime temperature of 25°C. All locusts were fed a diet of organic,
hydroponically grown wheat grass once a day with supplementary organic wheat germ available
55
a) b)
Figure 2.16 | Adult Locust Husbandry. Locust colony was maintained in a light and temperature-
controlled incubator. Locusts of different life cycles were housed in separate cages. Cages were
adapted to include a plexiglass front panel and locking door. Locusts were fed a diet of organic
wheat grass and organic wheat germ.
as needed. Once reaching sexual maturity, cups were placed into each cage for oviposition. Two
cups consisted of hydrated vermiculite and a third cup contained hydrated sand. Cups were left
in the cages for 2-3 days, at which point, they were replaced with freshly made cups, lidded, and
placed into a high-humidity incubator for hatching purposes (Figure 2.17). Each cup collecting
cycle continued for 3-weeks. The incubation time for these egg-rich cups was approximately 17
days, at which time, cups that contained hatchlings were placed into nursery cages and de-lidded.
Once locusts were large enough (3rd instar), cages were vacuumed every 2-3 days to remove any
frass, deceased individuals, and/or other debris.
56
a) b)
Figure 2.17 | Hatchling Locust Husbandry. Locust colony maintenance included ensuring
reproductive success with future generations. a) Cups filled with sand or vermiculite contain egg
sacs from which locust emerge. b) Hatchling cage one week after initial hatching depicts
numerous first and second instar locusts.
Honeybees
Honeybee cages were custom fabricated. Wooden back, side and top panels framed the cage
with #8 metallic screens covering the bottom. A pair of one-inch diameter holes were drilled into
the wood panel atop the cage and metallic screening was secured underneath. A front plexiglass
panel was installed to allow for adequate visibility (Figure 2.18). The honeybee incubator (Powers
Scientific, Inc.) was set to a 24-hour dark cycle at a constant temperature of 34.2°C. Two vials of
simple syrup (50% sucrose, 50% DI H2O) with small holes drilled in the lids for accessibility were
placed in the holes atop the wire mesh affixed to the top wood panel to allow bees constant
57
a) b)
Figure 2.18 | Honeybee Husbandry. Worker honeybees are housed in a dark, temperature- and
humidity-controlled incubator. Custom bee boxes were fabricated featuring a plexiglass front
panel for visualization. Bees were allotted a 50% sucrose solution and a pollen substitute as
needed.
a) b) c)
Figure 2.19 | Locust Exoskeleton Stabilization. a, b) Young adult locust stabilized on a clay
platform with plastic body and head supports. c) triangular pillars were constructed from batik
wax.
58
access to adequate nutrients. A vat of DI H2O was kept in the bottom of the incubator to maintain
a ~50% humidity level.
Surgery & Electrophysiology
Locusts
a) b)
c) d)
Figure 2.20 | Locust Exoskeleton Excision and Neural Surgery. A batik wax bowl was constructed
to stabilize and isolate the locust head. Prior to initial incision, the bowl was filled with a room
temperature locust saline to prevent tissue desiccation. The exoskeleton was excised using
microscissors and glandular tissue was removed with microtweezers. The brain was stabilized on
a wire platform and desheathed.
59
All neural recordings were conducted on post-fifth instar locusts (Schistocerca americana) of
either sex raised in a crowded colony. Locusts were initially immobilized on a surgical platform
(Figure 2.19a, b). Batik wax towers were constructed on either side of the head (Figure 2.19c).
Antennae were fed through 1/32” inner diameter polyethylene tubes and stabilized to the towers
using additional batik wax. A quick-acting, two-part epoxy was used to secure the antennae to
the inner walls of the polyethylene tubing. While the epoxy cured, a bowl of batik wax was built
to isolate the head region. The bowl was then filled with a room temperature, physiologically
balanced locust saline solution before removing exoskeleton and glandular tissue until the brain
was fully apparent (Figure 2.20). A drip tube supplying fresh saline was installed through the wax
bowl. Using super fine tip tweezers, the antennal lobes were desheathed following treatment
with protease.
a) b)
Figure 2.21 | Locust Electrode Insertion. Dual shank Neuronexus probe before and after
insertion. Required penetration depth is only 50-100 µm as the antennal lobe glomeruli are
positioned around the cortex of the lobe. Shanks were advanced slowly while watching for high
signal-to-noise ratios and action potentials on the Intan GUI.
60
Surgically prepared locusts were set inside a faraday cage to mitigate environmental
electrical noise. A 30-AWG silver reference wire coated with chloride ions was submerged into
the saline within the wax bowl. A 16-channel Neuronexus multi-electrode array was connected
to a 16-channel DIP socket and adjusted by a micromanipulator (WPI) for appropriate targeting.
The tips of the probes were placed directly atop the antennal lobe and slowly lowered (Figure
2.21) until neural spikes were clearly visible (~50-100 µm) on the Intan USB board interface GUI.
Odorants were then presented in a pseudorandomized order and neuronal responses were
recorded.
Honeybees
All honeybees were of foraging age and sourced from Arizona State University (Social Insect
Research Group, Smith Lab). The day prior to experiments, honeybees were collected from cages
within the incubator and placed into 50 mL conical vials. Six individuals were cryoanesthetized by
a) b) c)
d) e) f)
Figure 2.22 | Honeybee Harness and Surgical Preparation. Honeybees were placed inside a
custom-designed, 3D-printed harness and secured with dental wax.
61
inserting the conical vials into an ice bucket for ~1 minute until motion cessation. Using
entomological forceps, honeybees were removed from the vials and secured into a custom-
designed, 3-D printed harness by placing a piece of dental wax directly behind the head (Figure
2.22). Honeybees were left undisturbed for 30-minutes and then fed with a 50% sucrose solution
to satiation (Figure 2.23). All individuals were placed in a humidified cardboard box and left in
the dark overnight. The following day, restrained honeybees were given a small amount of the
Figure 2.23 | Honeybee Proboscis Extension Response. Honeybees were allotted simple syrup
the night before and morning of experiments. Here, a honeybee is engaging in a proboscis
extension response after having her antennae stimulated with the sugar water mixture.
62
Figure 2.24 | Surgically Prepared Honeybee. Honeybee has been prepared for neurosurgery by
affixing the antennae forward-facing with eicosane. The head capsule has also been shaved with
a microscalpel.
1:1 water-sucrose solution to check for proboscis extension response. Individuals that did not
present their proboscis were not used for electrophysiological experiments. Antennae were
maneuvered into a forward-facing orientation using a custom-designed antennae tool and
stabilized at the base of the scape and the pedicel with eicosane (Figure 2.24). The head was
shaved using a microscalpel and a window of the head capsule exoskeleton was excised (Figure
2.24). Any glandular tissue atop the brain was removed to allow for direct electrode access.
Honeybees were placed inside a faraday cage and a small piece foam was placed behind
the abdomen, to reduce body movement. A 30-AWG silver reference wire coated with chloride
63
ions was hooked inside the head capsule window. A four-channel twisted wire tetrode was
connected to the 16-channel DIP socket and adjusted by a micromanipulator for appropriate
a) b)
c) d)
Figure 2.25 | Honeybee Electrode Insertion. Custom made tetrodes have been inserted into the
antennal lobe to a depth of 100 µm. c) The twisted, 4-channel probe is clearly visible and the Ag-
Cl reference wire has been hooked underneath the head capsule cavity. d) A single drop of
honeybee saline solution is applied to prevent tissue desiccation during neural recordings.
64
targeting. The tetrode tip was brought to the antennal lobe surface. The tetrode was lowered
into the antennal lobe (Figure 2.25) until clear neuronal spikes were observed (<100 µm) on the
Intan USB board interface GUI. Adequate electrode targeting was confirmed by presenting a test
odorant and checking odor-evoked responses. Once electrode positioning was finalized, 20 µLs
of honeybee ringer solution was added to prevent tissue desiccation (Figure 2.25). Individuals
were left undisturbed for 10 minutes to allow for the brain to settle around the electrode.
Odorants were then presented in a pseudorandomized order and neuronal responses were
recorded.
Signal Analysis
Data Preprocessing
All analysis of neuronal signals was performed using MATLAB, R2021b (MATHWORKS, Inc.) and
IGOR-Pro 3.04. The MATLAB RHD file reader program, provided by Intan Technologies, was
adapted to enable highly efficient file reading and data processing. All data was high pass filtered
using a sixth order Butterworth filter with a cutoff frequency of 300 Hz. Stimulus presentation
periods were identified using an external signal delivered to the Intan USB Interface Board. If
necessary, data were aligned according to the stimulus presentation owing to the use of different
software programs with independent clock cycles. Data were saved in experiment-specific
folders, corresponding to individual electrode positions. At this point, all data underwent a spike
sorting procedure as well as root mean square transformation.
Spike Sorting
Data were converted to big-endian format in order to enable file importation into IGOR-Pro 3.04.
A supervised spike sorting program was used to identify spiking events based on experimentally
65
derived parameters. Covariance matrices were generated for waveforms that exceeded a
threshold between 2.5-3.5 times the standard deviation of the data. Sweep width and peak
position of identified super-threshold waveforms were chosen to ensure the entirety of the
waveform was captured. Common values for sweep width and peak position were 55 samples
and 25 samples, respectively. A model was generated based on the identified spiking events and
their time discrepancies that were seen in all four channels of the tetrodes. Sorted spike classes,
corresponding to potential putative neurons, were required to pass a number of tests before
further consideration. The distance (based on covariance) to a spike’s nearest neighboring spike
was considered. Additionally, the standard deviation within an individual spiking class and the
percentage of interspike intervals that were less than 20 ms were also considered. For locust
recordings, Neuronexus-based tetrodes, with an inter-electrode pitch of 25 µm, were sufficiently
separated in space to consider all three tests. Here, the nearest projection was required to be
greater than five standard deviations away and the percentage of inter-spike intervals less than
20 ms as well the intra-spike standard deviation could not be over 10%. For honeybee recordings,
custom-made twisted wire tetrodes did not provide enough spatial separation to apply the same
demands. In this case, the nearest projection was also set to a minimum of 5 standard deviations
away, but the intra-spike standard deviation was not considered Additionally, due to the spiking
nature of local neurons in the honeybee antennal lobe, the minimal percentage of inter-spike
intervals less than 20 ms was increased to 20%. Any spikes that did not meet these criteria for
locusts or honeybees did not undergo further consideration and were discarded. Raster plots
from the remaining spike classes of putative neurons were visually inspected to ensure adequate
inter-trial reliability and inter-stimulus reliability for pre-stimulation baseline periods. Those
66
spikes that passed this subsequent inspection were included in a model applied to all stimuli of
an individual experiment. The model identified similar spikes in other neural traces from the
same recording location across all stimuli. This process was repeated for all experiments,
adjusting IGOR parameters as necessary to ensure spike consistency and reliability. The spike
times of putative neurons were saved, converted to MAT files, and concatenated into master
tensors corresponding to each stimulus.
Root Mean Square (RMS) Transformation
Data that were not subjected to spike sorting were processed according to an unsupervised root
mean square transformation. Baseline values, considered to be voltage samples within 2 seconds
prior to the stimulus onset, was averaged and subtracted from the entire neural trace in order to
remove any bias due to electrode drift or signal-to-noise ratio variability. These baseline-
subtracted voltages were trimmed to the time window of interest and subjected to a 500-point
RMS transformation. Values that were less than 500 samples from either the beginning or end of
the trace were gradually tapered off according to the mean of the remaining samples. A 500-
point smoothing window was then applied to reduce any potential jitter due to natural biological
variability or thermal noise. Signals from all four tetrode channels were averaged together to get
an overall channel non-specific response. These RMS processed neural signals were
concatenated into master tensors corresponding to the appropriate stimuli.
Data Processing & Neural Population Analysis
Experiment-specific analyses were performed to visualize the response characteristics of an
individual electrode position or putative neuron. For raw signals, per-stimulus voltage traces
depicted signal-to-noise ratio and variability of responses based on the presented stimulus. For
67
RMS-processed data, per-stimulus voltage traces could also be visualized. For spike sorted data,
raster plots and peri-stimulus time histograms depicted possible stimulus-specific neuron
response patterns as well as inter-trial reliability of spiking events. More often, population-wide
responses were analyzed using the master tensors constructed from RMS-processed and spike
sorted data. For spike sorted data, the average of the 2 seconds prior to the stimulus were
subtracted from all experiments. For RMS-processed data, this was already performed during the
pre-processing regimen. Data were then binned according to the user-defined bin size. Master
tensors corresponding to individual stimuli were concatenated into one overall master tensor
with the following dimensions: number of neurons (for spike sorted data) or electrode positions
(for RMS-processed data) ´ number of binned samples ´ number of trials ´ number of stimuli.
Population peri-stimulus voltage traces or peri-stimulus time histograms with standard
error of the mean indicators were created to visualize population-averaged responses to each
stimulus. To further visualize population-wide response dynamics, principal component analysis
(PCA) was used as a dimensionality reduction technique. Here, data were projected to new
coordinates defined by the three largest eigenvectors of the dataset. This linear transformation
maps data to a new coordinate system defined by the directions of maximal variance within a
particular dataset and is routinely implemented to determine any potential underlying structure
within high-dimensional data. Owing to the temporal evolution of neural responses, data were
projected to this new PCA subspace and connected in a temporal fashion, forming neural
trajectories. These neural trajectories were smoothed slightly for visualization with a low-grade
cubic spline interpolation technique. Previous research has determined that these neural
trajectories correspond to stimulus-specific features, such as chemical identity and
68
concentration[516]. Additionally, linear discriminant analysis (LDA), was routinely used to
visualize datasets in an alternate dimensionally reduced subspace more suited to classification-
based problems. LDA differs from PCA in that it uses a priori class information to determine the
optimal transformative subspace. The algorithm does this by finding a linear transformation that
maximizes the between-class scatter while minimizing the within-class scatter of the projected
classes. Here, the data was initially normalized according to the mean of the entire dataset.
Means of all normalized data and of all stimulus-specific data were then calculated. The
eigenvectors of these resultant scatter matrices were computed, and all data was projected onto
the first three of these new component dimensions to form a new LDA subspace. An optional L2
regularization parameter was added to mitigate any potential overfitting of the algorithm when
considering overdetermined systems.
For quantitative analysis, data was handled solely in the original high-dimensional space.
Here, by using a leave-one-trial-out approach, training templates corresponding to each stimulus
were constructed. Test trials were held out and thus, not used for template construction.
Progressing through the temporal evolution of the neural signal, each sample was compared to
each training template using an L1 (Manhattan distance) or L2 (Euclidean distance) norm metric.
The training template that minimized this norm value, dictated classification of the test sample
bin. Cycling through all combinations of trials for template construction, enabled each trial to be
the held-out test trial exactly once. The classifier performance was then visualized by plotting the
presented odor versus the odor predicted by the model in a confusion matrix. Here, correct
classifications are indicated by entries along the main diagonal. Since insects rely heavily on both
spatial and temporal response characteristics for stimulus identification, we performed a winner-
69
take-all approach for a particular trial. This required taking the mode of all the bin predictions of
an individual trial as the appropriate predicted class. We constructed trial prediction-specific
confusion matrices for this subsequent approach. While the training templates and test trials are
independent in the leave-one-trial-out approach, we also formulated completely non-
overlapping train and test trial sets and performed similar norm calculations and confusion matrix
visualizations.
70
CHAPTER 3 | DISCRIMINATING CANCER BIOMARKERS USING A NOVEL INSECT
BIOSENSOR
Chemical Sensors
A key factor in the development of chemical sensing systems is the selection and integration of
optimal sensing materials. For each stimulus, sensors must undergo rigorous testing regimens to
determine whether they are effective for the application at hand. For example, a sensor designed
to aid in the diagnosis of bronchial asthma, in which elevated levels of exhaled nitric oxide are
present, must first and foremost be capable of detecting nitric oxide[1, 20-23]. Thus, the most
critical quality of an effective breath-based diagnostic is responsiveness to the chemical(s)
associated with the disease of interest. However, sensor response characteristics can be affected
by a host of different factors that may be inherent to the sample itself or the sampling
environment, complicating device reliability. One of the most prominent challenges for
manmade sensors, such as GC-MS and enoses, is the processing of stimuli characterized by high
levels of humidity. The presence of water vapor is well known to negatively affect sensor
detection limits. In contrast, biological olfactory systems, having evolved in the natural world,
may in fact benefit from the high water vapor concentration of breath samples[524, 525].
Another factor affecting chemical reaction kinetics is sensor operating temperature. Higher
temperatures increase the rate of chemical reactions and can shorten processing time associated
with manmade chemical sensors. Additionally, the shorter pre-sensing period mitigates the time
allotted for the breakdown of highly sensitive volatiles. This enhanced sensing rapidity can help
to maintain biomarker integrity and provide a more accurate profile of exhaled cellular
metabolites. Yet the heating of volatile components can itself encourage molecular degradation.
71
Both vertebrates and invertebrates can rapidly process chemical signals and identify stimuli
without the need for such highly controlled extrinsic thermal regulation[526, 527]. The
fundamental goal of biological olfaction is to encode stimulus-specific information in such a way
that reliably shapes perception and directs future behavior. As such, organisms must rapidly
encode chemical stimuli in unique manners that allows for maximal perceptual discrimination.
Canines, for example, have exquisite chemosensory abilities and are routinely employed as
chemical biosensors. A number of experiments have demonstrated that canine olfaction can be
used to diagnose pathological states using breath samples. To perform this feat, canines must be
able to identify subtle alterations in metabolic byproducts contained in breath profiles indicative
of the disease state.
Cellular Metabolism & Breath-Based Biomarkers
At any one time in the human body an assortment of up to 8700 possible biochemical processes
are ongoing[528]. Intermediary metabolic reactions are responsible for synthesizing simple
molecules or polymerizing them into complex macromolecules (anabolism), breaking down
compounds to release energy (catabolism), and removing any chemical byproducts (waste
removal). Chemically sensitive transporter proteins bind to and transport glucose molecules into
cells where they are phosphorylated and most often undergo glycolysis[529]. In the presence of
oxygen and two key substrate molecules, aerobic glycolysis converts glucose to the intermediate
metabolite pyruvate. Pyruvate is then converted to acetyl coenzyme A before entering the TCA
cycle to be converted into carbon dioxide, water, and adenosine triphosphate. In anaerobic
conditions, glycolysis converts glucose into lactate, which does not undergo further
oxidation[530]. Instead, the Cori cycle transports the lactate to the liver, which converts it back
72
to glucose via gluconeogenesis. The overarching goal of this intricate interplay of cellular
functions is to maintain homeostasis and the disruption of a single subprocess can lead to the
development of a particular disease state[531].
Diabetes mellitus, for example, is a disorder that is characterized by elongated periods of
hyperglycemia. Here, an enzymatic imbalance enhances gluconeogenesis while decreasing the
rate of glycolysis. This in turn triggers excessive hepatic glucose production and high
concentrations of glucose in the blood for extended periods of time[532]. The inability to
effectively transport glucose into the cell alters feedforward and feedback regulatory pathways.
Anabolic and catabolic dysregulation can exacerbate bouts of hyperglycemia inducing significant
levels of oxidative stress[533-536]. Carbon dioxide, digested nutrients and metabolic waste are
transported to the lungs via the pulmonary circulatory system. Here, carbon dioxide and other
highly volatile chemicals can diffuse across the blood-air barrier into the alveoli of the lungs, and
possibly airways, and be excreted in the breath[537]. The chemical profile of exhaled breath
contains a substantial degree of information on underlying metabolic processes and the
existence of particular pathologies can be inferred. For diabetics in particular, acetone
production has long been known to increase during diabetic ketoacidosis[25, 538, 539] and
breath-based concentrations correlate with glucose serum levels[540]. This allows for the
development of effective breath tests to measure acetone concentrations, which are widespread
in research and gaining traction for clinical implementation[26, 541-549].
Unlike acetone as a marker for diabetes, a single volatile biomarker indicative of cancer
has not yet been discovered and is highly unlikely. Rather, complex patterns of specific VOC
concentration increases and decreases have been observed in cancer-related breath testing
73
research. The subtle changes in numerous breath constituents are suspected to reflect the
underlying heterogeneity and multifarious nature of the disease. The molecular mechanisms of
cancer have been studied ad nauseum for over two and a half centuries. Cancer cells deregulate
natural metabolic signals for achieving homeostasis in favor of those promoting chronic
proliferation. Warburg first demonstrated that even in oxygenated environments, most cancer
cells preferentially engage in glycolysis but not subsequent oxidative phosphorylation, as often
occurs in healthy cells[550]. The intermediary pyruvate molecule is converted to lactate and
either secreted into the extracellular milieu or recycled for cell-specific nutrients[551, 552].
Inefficient energy production of aerobic glycolysis is compensated by upregulating glucose
transporter proteins, enabling significantly more glucose to enter the cell[553]. The combination
of the unfavorable energy output of glycolysis and the introduction of additional glucose
transporter proteins into the cell membrane, causes the cell to deplete glucose from surrounding
tissues[554]. Moreover, the elevated levels of lactate in the extracellular space promote
immunosuppression, thereby preventing tumor cell recognition and enabling further
proliferation[552, 555, 556]. A host of downstream effects are common including the synthesis
of excessive fatty acids, tumor formation and acidification of the local cellular environment[557].
This leads to the aggregation of metabolic waste products at levels not seen in healthy conditions,
which, if volatilized and excreted via the breath, can be indicative of underlying pathology.
Numerous studies have investigated the efficacy of breath-based technologies for the diagnosis
of lung, breast, and other types of cancers. Blood- and breath-based concentrations of certain
aldehydes, alcohols, and ketones have been shown to differ based on the underlying disease as
well as the affected organ[558].
74
Locust Olfaction
The desert locust (Schistocerca gregaria) is a widely used insect model for exploring biologically
relevant olfactory processing schemes. Preliminary work by Laurent and others, elucidated
several critical aspects locusts employ for odorant discrimination, including local signal
processing networks and sparse signal transmission to higher neural regions[359, 360, 511, 516,
518, 559, 560]. The exact dynamics of these networks remain unknown but theoretical and
computational models continue to elucidate key elements of functional schemes[514, 560-563].
Locust antennae display a variety of disparate chemosensory sensilla dedicated to olfaction.
These sensilla house olfactory receptor neurons (ORNs), which are responsible for chemical
recognition and signal transduction. For example, basiconic sensilla house anywhere from 30-50
ORNs, whereas trichoid sensilla house just three[480]. The majority of ORNs are broadly tuned,
responding to a number of different volatile chemicals based on specific stereochemical
properties. This enables a highly complex and competitive ligand binding environment within
each sensillum, especially those with greater numbers of similarly tuned ORNs. Relative to other
insects, locusts have been found to display unique anatomical features. Individual ORNs branch
to and innervate several glomeruli within the antennal lobe, suggesting a highly adaptable
network with extensive signal processing capabilities[564]. Interestingly, unlike the fewer (50-
100), large glomeruli found in other insects, the antennal lobe of locusts contains 1000
microglomeruli[355, 564]. These highly interconnected glomeruli are positioned near the surface
of the antennal lobe and arranged around a central fiber core (Figure 3.1) [515]. This dense fiber
75
a) b)
c)
Figure 3.1 | Locust neuroanatomy. a) Schematic drawing of locust brain with emphasis on
olfactory pathway. Olfactory receptor neurons project axons to the antennal lobe via the
antennal nerve (ant n.). A projection neuron (labeled as NOTG I) is depicted. Locust projection
neurons can receive signals from eight or more glomeruli, eliciting spikes with sufficient
concurrent input activity. The axonal projections innervate Kenyon cells of the mushroom body
(a.c.) in a sparse fashion. Reproduced from [564]. b) Microglomeruli can be visualized with a DAPI
stain, confirming their presence around the cortical region of the antennal lobe. Reproduced
from [565]. c) An exposed locust brain following surgery procedure. The antennal nerves,
antennal lobes and mushroom bodies are clearly visible.
76
network efficiently converges input signals from a total of 50,000 ORNs onto a mere 300
inhibitory local neurons. Lacking axons, local neurons utilize dendrodendritic graded electrical
potentials, instead of spike-based action potentials, to aid in local signal conditioning. The neural
network within the antennal lobe is an immensely powerful signal processor, capable of
improving signal-to-noise ratios and enabling background-invariant odor coding[311]. While
some of these local neurons extend to only a few glomeruli, others arborize across the majority,
if not all, of the antennal lobe neural circuitry. After signal conditioning, an assortment of 830
projection neurons, which receive input from multiple different glomeruli, transmits the signal
a) b)
Figure 3.2 | Antennal lobe-mushroom body signal sparsening. a) Four projection neurons elicit
unique bursting response properties to the same and different odorants. These neurons display
a considerable amount of spontaneous baseline activity before the stimulus arrives (gray and
black bars). The responses of these neurons are mapped to and visualized in a lower two-
dimensional space as a cloud of points. b) Kenyon cells of the mushroom body exhibit extremely
sparse firing activity and little to no background spiking. Note that there is a significant of
divergence between the projection neurons (830) to Kenyon cells (50,000), suggesting
considerable specificity. The signal transformation that occurs between projection neurons and
Kenyon cells improves separability in this lower two-dimensional space. Reproduced from [566].
77
to higher order neural regions, such as the mushroom body. Here, 50,000 highly specific Kenyon
cells exhibit little to no spontaneous activity and only fire one to two spikes when activated
(Figure 3.2) [566]. Each of these Kenyon cells is contacted by approximately 50% of the PNs,
providing a highly intricate, multidimensional input signal for each neuron[567]. Receiving a
significant number of inputs, Kenyon cells function as coincidence detectors, only firing when a
large portion of the afferent projection neurons all fire simultaneously. This decorrelating
function of the mushroom body can reduce the complexity of the incoming signal, making it far
more interpretable while maintaining a substantial encoding state space[560, 568]. The response
dynamics of mushroom body neurons play a vital role in learning as well as memory formation,
consolidation, and retrieval via feedback mechanisms.
Locust-Based Cancer Biomarker Differentiation
Previous work has demonstrated that locusts can detect and differentiate between various
alcohols, aldehydes[516, 517, 519], and even explosive chemicals[523]. In order to determine the
feasibility of using the locust olfactory system as a breath-based cancer diagnostic, we sought to
test the responsiveness of projection neurons to a number of volatiles found to be significantly
upregulated in lung and breast cancer[78, 132, 569]. Moreover, we intended to demonstrate that
the chemicals could elicit unique and reliable responses capable of providing successful stimulus
discrimination. For the lung cancer volatile chemical panel, we used decane, 2-methylheptane,
6-6-4-4 pentamethylheptane, propylbenzene and undecane. For the breast cancer volatile
chemical panel, we used hexanal, nonanal, pentanal, trichlorotheylene and undecane. All
odorants were diluted in paraffin oil (1% v/v) and stored in 20 mL glass vials with 1/32” diameter
78
Lung Cancer Breast Cancer
Decane 0.67 ppm Hexanal 0.43 ppm
Methylheptane 3.14 ppm Nonanal 0.2 ppm
Pentamethylheptane 0.83 ppm Pentanal 10.8 ppm
Propylbenzene 1.61 ppm Trichloroethylene 53.23 ppm
Undecane 0.22 ppm Undecane 0.22 ppm
Table 3.1 | Biomarker stimulus panel. All odorants were chosen due to their identification as
breath-based volatile biomarkers in lung and breast cancer. A large number of volatiles have been
linked to cancer with significant variation between studies. Note that some chemicals may be
implicated in more than one type of cancer, such as undecane seen here.
polytetrafluoroethylene (PTFE) tubing serving as inlet and outlet lines. For additional information
on odorant preparation, see Chapter 2: Methodology- Stimulus Creation. Due to differences in
volatilities, Raoult’s law was used to approximate the gaseous concentrations (Table 3.1).
The brain of locusts was exposed and perfused with a custom-made locust saline
throughout the duration of the experiment to prevent tissue desiccation. Microelectrode arrays
Figure 3.3 | Experimental design schematic. The air cylinder passes zero-contaminant air into
the olfactometer, which controlled flow trajectories using a series of solenoid valves. Upon
stimulus delivery a final valve (not shown) redirected the airstream in-line with the vial to the
insect antenna. For more information see Chapter 2: Methodology- Stimulus Delivery. Here,
odor delivery demarcated by the gray box elicits a sharp transient response from locust
projection neurons. The neuron quickly moves into a steady state response before a return to
baseline upon stimulus offset.
79
were inserted into the antennal lobe until spikes were visible and purified odorants were
presented in a pseudorandomized fashion (Figure 3.3). For more information, see Chapter 2:
Methodology- Stimulus Presentation. The locust olfactory system appeared highly responsive to
many of the presented odors. Though some stimuli, such as hexanal, seemed to generate
distinctive responses in most neurons, others appeared to preferentially activate more select
Figure 3.4 | Biomarker peristimulus voltage traces. Neural traces from a representative neural
recording location demonstrate unique spatiotemporal responses to different odorants. Four
second stimulus presentation indicated by gray box. Spontaneous spiking occurred prior to
stimulus onset as is expected from projection neurons. Most odors caused a significant increase
in spike rates, especially during the transient response period. Some odors, such as hexanal and
pentanal, display a unique spiking response to the stimulus onset as well as a separate pattern
to the stimulus offset. Note that in some of the traces, spikes of different amplitudes are
discernible, suggesting the presence of multiple neurons from this recording location.
80
neuronal subsets. In order to investigate the quality of the data, we initially observed individual
recordings and odor-evoked response dynamics. Neural signals recorded from an individual
electrode elicited reliable temporal responses over all five trials. Figure 3.4 depicts voltage traces
after pre-processing with a 300-Hz high-pass filter to remove low frequency signals associated
with local field potentials. Increased neural firing rates indicate that neuron(s) in this particular
electrode recording vicinity were able to detect most of the volatile molecules included in our
cancer odor panel. The broad tuning curves of many projection neurons in the locust antennal
lobe suggests that each neuron may respond to a particular feature of the presented stimulus.
On this basis, we recorded from 24 different electrode positions across nine individual locusts.
Evidence suggests that the antennal lobe and associated neural circuitry is largely conserved
within species. Thus, each recording was considered independent as different areas of the
antennal lobe were targeted. To investigate the population-wide responses from all recordings,
we first employed a spike sorting procedure (see Chapter 2: Methodology- Signal Analysis). Spike
times corresponding to putative neurons were recorded. Owing to the 20-Hz oscillatory
integration cycles observed in the antennal lobe and the corresponding spiking activity in Kenyon
cells of the mushroom body, we considered the total number of spikes over a 50 msec window.
A master tensor was created consisting of spikes per bin ´ trials ´ number of bins.
A population-wide peri-stimulus time histogram was constructed by plotting the average
value across all binned neuron-specific responses for a given odor (Figure 3.5). While these peri-
stimulus time histograms demonstrated unique response properties of all neurons as an
averaged group, more effective analyses were performed to further investigate the underlying
structure of the data. Various time windows were considered to determine the response period
81
a)
b) c)
Figure 3.5 | Population-wide peri-stimulus time histogram demonstrates unique response
trajectories. a) Neural traces are plotted from one second prior to stimulus onset to four seconds
after stimulus offset, as response dynamics can continue well after stimulus cessation. Unique
response patterns were observed with the transient period immediately after stimulus onset
eliciting the most significant odor-evoked activity in response to all odors. b) On responses were
seen approximately 300 msecs after stimulus delivery. This agrees with the latency for
82
Figure 3.5 (cont’d)
intracellular recordings and other insects. c) Some odors generated relatively informative off
responses upon stimulus cessation, though, on average, they were less stark than the transient
on responses. Note in all plots, mean averages are depicted with a dark colored line and the
standard error of the mean over all trials is plotted as a semi-translucent colored patch
surrounding the mean average.
capable of issuing optimal stimulus separability and classification. Linear dimensionality
reduction techniques were utilized for data visualization purposes. Data was processed according
to principal component analysis (PCA) and resultant neural trajectories were plotted as functions
of time for the three dimensions explaining maximal variance across the dataset. For data
processing specifications, see Chapter 2: Methodology- Signal Analysis. Figure 3.6 depicts neural
trajectories for the four seconds following stimulus onset (Figure 3.6a), the two seconds
following stimulus onset (Figure 3.6b), and the one second following stimulus onset (Figure 3.6c).
For all analyzed time periods, trajectories are seen to move to different regions of the PCA
subspace during the transient response period before returning to near baseline upon reaching
a steady state. This suggests neurons exhibit unique, stimulus-specific response dynamics upon
a) b) c)
Figure 3.6 | Neural trajectories exhibit unique temporal dynamics after being projected into
principal component space. The first bin of all trajectories are zeroed so that all traces begin at
the same point in PCA space. Subsequent bins are subtracted by the initial value to maintain
proper response dynamics. Neural trajectories evolve in specific manners for a) the four second
period after stimulus onset, b) the two second period after stimulus onset, and c) the one second
period after stimulus onset. Total number of neurons: n = 45.
83
a) b) c)
Figure 3.7 | Stimulus-specific clusters consisting of activity of individual time bins separate from
one another in LDA space. Distinct groups begin to emerge by processing data according to a
linear discriminant analysis transformation and projecting points corresponding to the activity of
individual bins into this new subspace. This supervised dimensionality reduction technique was
performed for a) the four seconds, b) two seconds, and c) one second after stimulus onset. Total
number of neurons: n = 45.
stimulus presentation, especially during this initial transient period. To further investigate the
underlying structure of the dataset, an alternative linear dimensionality reduction algorithm,
linear discriminant analysis (LDA), was performed. This supervised technique serves to maximize
the variance between different groups, while minimizing that within a particular group. Again,
time periods of the four seconds (Figure 3.7a), two seconds (Figure 3.7b), and one second (Figure
3.7c) after stimulus onset were considered. Each dot corresponds to the activity of an individual
time bin. As such, for longer time windows and smaller bin sizes, more dots were generated, and
graphs became somewhat polluted. Nevertheless, the existence of stimulus-specific clusters
began to emerge in this new LDA hyperspace. Dimensionality reduction techniques such as PCA
and LDA are advantageous in their ability to present data in more interpretable manners.
However, depending on the complexity of the dataset, a significant loss in overall information
can occur.
For quantitative analysis, therefore, we generated stimulus-specific training templates
using a leave-one-trial-out cross validation methodology in the original high dimensional
encoding space. Training templates were formed from four out of the five trials, while holding
84
the final trial out as a novel testing trial. Since our spike sorting algorithm was able to extract 45
putative neurons, each training template and testing trial consisted of a vector of 45 different
values, each corresponding to an individual time bin over all suspected neurons. The stimulus-
specific training templates were averaged over the time window of interest and a Euclidean norm
metric was used to calculate the distance between each training template and individual bins for
the testing trials. The training template that minimized the norm distance to the bin of the testing
trial was considered to be the predicted class. For more details on our leave-one-trial-out
classification approach, see Chapter 2: Methodology- Signal Analysis. Resultant predictions were
plotted against the true values via a confusion matrix (Figure 3.8a). This confusion matrix depicts
the overall accuracy of classifying individual bins. While this allows for an expansive high
dimensional encoding space, it treats each time bin as an independent unit. In contrast, biological
systems can aggregate response information collected over a number of time bins until
a) b)
Figure 3.8 | Euclidian-norm based classifier performance. Here, classifier accuracies are
displayed as confusion matrices. High classifier accuracy is indicated by dark colors along the
main diagonal. The leave-on-trial-out train-test protocol was utilized in this scenario to train a
linear classifier. To assign predicted classes for the test trial, we selected the training template
that minimized the Euclidean distance (L2). Results are shown for a) individual bins for each test
trial and b) entire trials in a winner-take-all approach. Note in a the majority of test bins were
classified correctly and in b the classifier achieved 100% accuracy.
85
a) b)
Figure 3.9 | Manhattan-norm based classifier performance. Similar to figure 3.8, classifier
accuracies are displayed as confusion matrices. High classifier accuracy is indicated by dark colors
along the main diagonal. The leave-on-trial-out train-test protocol was utilized in this scenario to
train a linear classifier. To assign predicted classes for the test trial, we selected the training
template that minimized the Manhattan distance (L1). Results are shown for a) individual bins for
each test trial and b) entire trials in a winner-take-all approach. A different norm metric applied
here improved the bin-wise classification and did not affect the perfect performance seen
previously in trial-wise classification.
sufficiently confident to predict stimulus identity. Therefore, we employed a winner-take-all
approach where each trial in its entirety was classified, instead of individual time bins. This
technique showed a marked increase in prediction accuracy evidenced by the associated
confusion matrix (Figure 3.8b). We also sought to determine classification effectiveness using the
Manhattan norm distance (Figure 3.9) and maximum norm distance (Figure 3.10).
Since locusts routinely integrate signals within a 20-Hz oscillatory cycle, our default bin
size was 50 msecs. However, this time period may be an adaptation necessary for the rapid
decision making within highly dynamic natural environments. The inclusion of more time
windows, hence more oscillatory cycles, has been shown to be important for discriminating
between stimuli with a higher degree of similarity[570]. While the inclusion of single odorants in
our panel reduced the chemical complexity of the stimulus, these volatiles may not be ubiquitous
in locusts’ natural environment and classification accuracy may be highly dependent on allotted
86
a) b)
Figure 3.10 | Maximum-norm based classifier performance. Similar to figure 3.8, classifier
accuracies are displayed as confusion matrices. High classifier accuracy is indicated by dark colors
along the main diagonal. The leave-on-trial-out train-test protocol was utilized in this scenario to
train a linear classifier. To assign predicted classes for the test trial, we selected the training
template that minimized the infinity-norm (L¥) distance. In this case, only the absolute value of
the largest component of the vector is taken into account. Results are shown for a) individual
bins for each test trial and b) entire trials in a winner-take-all approach. In this scenario, the
classifier did not perform as well relative to using the Euclidean and Manhattan norms as distance
metrics. Moreover, we see our first misclassification in the trial-wise confusion matrix.
processing time. Therefore, we tested alternate bin sizes of 100, 150, and 200 msecs. Peri-
stimulus time histograms were characterized by smoother features with increasing bin sizes
(Figure 3.11). This is to be expected with a larger number of included data points, yet, critically,
a) b) c)
Figure 3.11 | Bin-size selection affects peri-stimulus time histogram dynamics. The graphs
indicate the population average response for all neurons from one second prior to stimulus onset
to four seconds after stimulus offset. The stimulus presentation period is demarcated by the light
gray box. Different bin sizes were considered to examine their effect on response dynamics. a)
100-, b) 150-, and c) 200-msec bin sizes are shown. As the bin size increases, the data becomes
smoother, reducing fine response dynamics that could be a result of noise or function as key
information for stimulus classification.
87
a) b) c)
Figure 3.12 | Bin-size selection affects neural trajectories in PCA space. Data was plotted after
processing via principal component analysis. As expected from the peri-stimulus time histograms,
the minor voltage fluctuations evident in the a) 100-msec scenario were seen slightly with b) 150-
msec and not at all with c) 200-msec bin sizes. In all three cases, unique neural trajectories are
observed indicating that different odorants induce variations in temporal response patterns.
high consistency was observed in the overall response patterns regardless of bin size. All three
bin sizes were processed via principal component analysis (Figure 3.12), linear discriminant
analysis (Figure 3.13), and tested using the previously used leave-one-trial-out methodology
(Figure 3.14).
Spike sorting has the benefit of creating sparse matrices, significantly reducing
computational complexity and overhead. Yet, the employed spike sorting algorithm used here
was a supervised approach, requiring careful experimenter intervention and evaluation of each
spike class as belonging to a viable putative neuron. Effective sensors, apart from permitting
a) b) c)
Figure 3.13 | Bin-size selection affects neural clusters in LDA space. Neural responses were
processed using a linear discriminant analysis technique and the time bins were plotted in LDA
space. Here we can clearly see fewer time bins with a larger bin size. a) With 100-msec bin
sizes, stimulus-specific groups were beginning to emerge, but some still remained clustered
together. b) 150-msec bins enabled well-separated clusters and c) 200-msec bins enabled
clusters to occupy discrete points within the LDA subspace.
88
a) b) c)
d) e) f)
Figure 3.14 | Bin-size selection quantified via confusion matrices. We subjected data of different
bin sizes to a leave-one-trial-out train-test procedure. We could visualize the accurate predictions
and which odorants induced more similar neural responses via confusion matrices. Bin-wise
classification results are shown for a) 100-, b) 150-, and c) 200-msec bin sizes. With increasing
bin sizes the accuracy improves, though the classifier is also predicting fewer time bins. The trial-
wise accuracies for d) 100-, e) 150-, and f) 200-msec bins all reached 100%. All classifier
predictions were made using the Euclidean norm.
adequate classification performance, ought to enable rapid readout and minimal supervised
intervention. Therefore, we performed a relatively simple, fully automated root-mean-square
transformation capable of reducing the total number of data points while retaining the intricacies
of spike response patterns (Figure 3.15). A number of different point filters were tested to
determine the minimal resolution necessary to retain unique stimulus-specific response patterns.
Processing the data with a 500-point filter seemed to over-smooth the data, with the dissolution
of potentially important signal elements. The 100-point filter retained a large number of these
features, yet some of the minute voltage changes may reflect natural biological response
variability and would not enhance future unknown stimulus classification. All RMS processing was
conducted with a series of two 500-point filters, one for applying a smoothing algorithm and
89
a) b)
Figure 3.15 | Root-mean square transformation preserves signal integrity. In an effort to move
towards real-time analysis and readout, data was initially smoothed and then subjected to a
moving point RMS filter. For details on signal transformation, see Chapter 2 Methodology- Signal
Analysis. a) Data from all four tetrode channels was efficiently combined and transformed from
20,000 samples per second to 20, while retaining signal integrity. b) A number of different point
filter values were considered. For larger point filters (top), some fine neural response features
were removed, while for smaller point filters (bottom), many of these were retained.
another for the root-mean-square transformation. For additional details, see Chapter 2:
Methodology- Signal Analysis. We performed similar analytical techniques on the RMS-
processed dataset, that we had utilized for the previous spike sorted data. Peri-stimulus voltage
traces demonstrated clear population-wide neural responses unique to the presented odorant
(Figure 3.16). A more salient off response was observed for a number of the stimuli relative to
the spike sorted data, suggesting that response dynamics of these neurons were not able to be
captured by use of the spike sorting algorithm. Since the RMS analysis incorporates the total
energy from the recording location, it may retain pertinent information that would otherwise be
lost in spike sorting putative neurons. We then performed principal component analysis (Figure
3.17a) and linear discriminant analysis (Figure 3.17b) to aid in data visualization. Similar to spike
sorted data visualization results, neural trajectories and stimulus-specific clusters began to
emerge in the PCA and LDA subspaces, respectively. A total of 24 recording locations defined the
90
a)
b) c)
Figure 3.16 | Peri-stimulus voltage trace of RMS-processed data. Intricate neural response
dynamics can be observed from population-wide averages of RMS-processed data. Here, clear b)
ON and c) OFF responses are apparent. This may be a result of retaining more stimulus-specific
information from the total energy signal of the recording location that was previously filtered out
during spike sorting.
91
a) b)
Figure 3.17 | Unique population dynamics observed in PCA and LDA subspaces. Here, we
performed similar dimensionality reduction algorithms as before, namely principal component
analysis and linear discriminant analysis. We kept only the three most prominent dimensions in
order to effectively analyze the underlying structure. The RMS-processed data generated a)
uniquely evolving neural trajectories in PCA space and b) stimulus specific clusters in LDA space.
Figure 3.18 | Confusion matrices for RMS-processed data. To quantify the performance of a
relatively simple linear classifier in 24-dimensional space, we used the leave-one-trial-out
training-testing procedure. Results were comparable to those attained by spike sorting. a, c) Bin-
wise and trial-wise based classification using the Euclidean norm as the distance metric. b, d) Bin-
wise and trial-wise accuracy using the Manhattan norm as the distance metric. Similar to the
spike sorting data, the Manhattan norm produced slightly better results that using the Euclidean
norm metric. The rapid processing and readout is a step towards real-time and completely
unsupervised analysis.
92
encoding state space, in which training templates were generated using the leave-one-trial-out
cross validation technique. This high dimensional classification in a 24-dimensional RMS-based
encoding space offered similar performance to the 45-dimensional space of spike sorted data
(Figure 3.18). Moreover, signal processing did not require experimenter supervision and total
processing time was drastically reduced.
Outlook
Biological olfaction is an immensely powerful chemical processing system that has evolved over
millions of years to sense a broad range of chemicals at minute concentrations. On the contrary,
manmade gas sensing for diagnostic capabilities is a relatively new field. These sensors display
extraordinary potential but, struggle to achieve the broad sensing capabilities and low parts-per-
trillion to parts-per-quintillion detection limits seen in biological olfactory systems. Moreover,
olfaction has evolved to perform background invariant stimulus identification, an extremely
important quality for tasks in which natural variability is expected.
This study represents a crucial first step towards determining whether insect olfactory
systems could be suitable as a medical diagnostic technology. Our work has demonstrated that
the locust olfactory system can be readily utilized to detect a number of key volatiles present in
the exhaled breath of cancer patients. It is important for a cross-reactive chemical sensing system
to display unique encoding of various chemicals or molecular features. Recognition of individual
features dramatically enhances the capacity of the system—an essential characteristic for any
device designed to characterize highly heterogeneous stimulus mixtures adequately and reliably.
We have also proven that stimuli containing unique chemical structures elicit stereotypical
responses in projection neurons of the locust antennal lobe and these spatiotemporal readouts
93
can be used to classify stimulus identity with an extremely high degree of accuracy. Even odorants
with similar chemical structures, such as hexanal and pentanal, were distinguished from one
another. These two volatiles have also been identified as chemical biomarkers of maladaptive
metabolic states, such as oxidative stress and lipid peroxidation. While our goal is to test the
efficacy of the locust olfactory system as a non-invasive cancer diagnostic specifically, it may also
have use as a broad-spectrum screening tool. Processes such as oxidative stress occur in diseased
populations as well as healthy individuals and is likely a natural phenomenon indicative of aging.
Depending on detection limits and disease separability, the locust olfactory system could offer a
modality by which to investigate underlying maladaptive metabolic processes that are not
necessarily considered to be pathogenic.
94
CHAPTER 4 | CANCER CELL LINE DIFFERENTIATION BY AN INSECT CHEMICAL
BIOSENSOR
Cancer Metabolism
Cancer is a highly complex and multifaceted disease consisting of hundreds or thousands of
different genotypes[571]. While some key features can be identified, different types and sub-
types of cancers can lead to significantly different biomolecular reactions. The Warburg effect of
aerobic glycolysis is considered to be a metabolic hallmark of cancer. However, the ubiquity of
this metabolic pathway has been questioned relentlessly since its original proposition. While
some evidence suggests the existence of damaged mitochondria[572, 573], others point to
decreased mitochondrial activity triggered by upregulations in specific transcription factors[574-
576]. Ultimately, given the rapid speed of glycolysis relative to that of oxidative
phosphorylation[577, 578], the mitochondria cannot process the plethora of resultant pyruvate
molecules. While these two models seem contradictory, different cancers may, in fact, be
characterized by these distinct features due to the impact of genetic variation or environmental
influences. This could lead to alterations in subsequent downstream molecular processes and, in
part, explain the subtle variations in biomarkers associated with different types of cancers.
Disease variability also stems from the identity of the affected organ and even the region within
said organ. Tumor formation is highly dependent on the local cellular environment, which is
shaped in large part by the excessive release of lactate by cancerous cells into the extracellular
milieu. Extracellular signals can be dictated by lactate-specific chemical reactions, which are
differentially impacted by variables such as the efficacy of the nutrient delivery system and the
distance to surrounding blood vessels[579-582]. The multifarious nature of cancer emphasizes
95
the pressing need to consider type- and subtype-specificity for effective diagnosis, prognosis, and
treatment.
Volatile-Based Cancer Detection
Breath-based chemical sensors have proved to be an enticing option for cancer detection.
Groundbreaking experiments over the past three decades have shown that the excretion of
trace-level volatiles differs between healthy and diseased individuals[583]. Generic volatile
biomarkers representative of oxidative stress and lipid peroxidation, such as ethane and pentane,
are often observed in higher quantities for those with underlying illnesses. However, the real
power of breath-based gas sensing is due to the subtle changes in other volatile metabolites that
are more indicative of certain types and subtypes of diseases[114, 140, 265, 584]. One study has
demonstrated that breath concentrations of several compounds not only differ between breast
cancer patients and healthy controls, but also among those with genetically determined breast-
cancer subtypes[134]. Here, GC-MS and electronic nose-based analyses were fairly accurate in
distinguishing group identity based on breath profiles alone. The ability to discriminate between
different disease states necessitates that the underlying biochemical processes related to a
specific disease state will elicit a unique assortment of volatiles, indicative of those processes.
Furthermore, accurate diagnosis requires sufficiently sensitive gas sensing technology to detect
at least some of these volatiles and discern an appropriate combination that allows for group
separability based on underlying pathology.
In vitro cell culture models are commonly used to test the efficacy of gas-sensing
technologies for volatolomics research[129-131, 137, 138, 266, 583, 585-587]. For volatile
headspace sampling, either an adsorbent material can be placed in the sealed flask or direct
96
headspace can be passed to the gas sensor. In the sorbent-based technique, metabolites
volatilize out of the cell culture medium and are absorbed by the sorbent trap. Subsequent
thermal desorption can be used to increase molecular energy and break chemical-sorbent bonds.
A high concentration of chemicals with relatively similar volatilities will be released at any one
time. In order to maximize the number of chemicals released from the trap, especially those with
lower volatilities, the temperature is incrementally increased over time. For technologies, such
as gas chromatography-mass spectrometry and ion mobility spectrometry, this is perfectly
appropriate as each chemical is considered independently of others in the mixture. However, for
electronic noses and biological olfaction, which create breathprint templates for entire samples,
the temporal nature of odorant delivery is critical. Passing the volatile headspace directly,
ensures a more homogeneous stimulus between individual presentations.
Having demonstrated biosensor responsiveness to and discriminability of various cancer
VOC biomarkers, our next set of experiments were designed to determine the ability of the locust
olfactory system to effectively differentiate between the volatile headspace profiles of cancerous
and non-cancerous cell lines. Evidence suggests that there are more than 500 volatiles in a single
breath sample, though the actual number is likely much higher as volatile identification is a
function of technological sensitivity[134]. Of these identified volatiles, approximately 10% appear
to be universal[36]. As such, an effective breath-based sensor must be able to process highly
complex mixtures with overlapping constituent chemicals. Additionally, considering the subtle
changes in breath biomarkers brought about by disease, detecting concentration differences
requires exceptional chemical sensitivity and highly effective pattern recognition capabilities. In
addition to differentiating between cancer and non-cancer samples, we sought to test whether
97
our forward-engineering biological-based approach was powerful enough to distinguish between
oral cancer subtypes. The cell lines chosen were based on preliminary data demonstrating that
the cancer cell lines (Ca922, HSC-3, and SAS) exhibit functional increases in both glycolysis and
oxidative phosphorylation relative to the non-cancer cell line (HaCaT). All cell lines were grown
Figure 4.1: Electrophysiological recording setup. a) Schematic of the VOC delivery and in vivo
neural recording setup. Cancer and non-cancer cell lines were cultured and placed inside airtight
flasks. The culture medium was the same for all cell lines. Emitted VOCs from the cell cultures
were sampled periodically by injecting a fixed amount of clean air into the closed flask using an
olfactometer. The duration and volume of cell culture VOCs delivered to the locust antenna were
controlled by the odor delivery setup. Extracellular neural recordings were obtained from the
locust antennal lobe before, during, and after odor delivery. Total airflow to the antenna was
kept constant throughout the experiment and delivered VOCs were removed quickly by an
exhaust placed behind the locust antenna. A raw voltage response of a neural recording is shown
for a 4 s long odor pulse. b) Image depicting the olfactometer unit connected to an in-line cell
culture flask. The final valve, with exhaust and stimulus flow lines, is shown in the lower right
portion of the image. c) Experimental setup depicting an in vivo locust preparation housed inside
a faraday cage. The brain has been implanted with a microelectrode array and an Ag-Cl reference
electrode completes the electrical circuit. The stimulus flow line passes either zero contaminant
clean air or air containing volatiles from the cell culture headspace to the locust antenna,
depending on stimulation parameters.
98
in identical conditions to ensure background VOC consistency, a critical component for any in
vitro volatolomics study[588]. For precise culturing conditions see Chapter 2: Methodology-
Stimulus Creation. Stimuli consisting of headspace volatiles were systematically passed to locust
antennae over a period of four seconds. Stimulus presentation was repeated five times with an
inter-stimulus interval of 60 seconds in order to ensure responses remained consistent over time
and to enable accurate sensor calibration. Microelectrodes were used to record neuronal action
potentials from projection neurons within the antennal lobe (Figure 4.1). Prior to each
experiment, ten pictures were taken from each flask to investigate cell health and density (Figure
4.2). Cells in each of these pictures were manually counted using ImageJ and longitudinal growth
curves were constructed (Figure 4.3).
Locust-Based Cancer Biosensor
We began by investigating odor-evoked individual projection neuron (PN) responses in the locust
antennal lobe to the volatile headspace of cancer and non-cancer cells. Cell culture VOC samples
were examined at 24 h intervals by in vivo PN recordings. Additionally, we used two control
odorants, hexanal and undecane, which have been implicated in earlier studies as putative cancer
biomarkers[589].
We observed VOC-evoked changes in neural spiking responses in most of the PNs
recorded. Since PNs are broadly selective to several odor stimuli and respond to specific odorants
or odor mixtures with distinctive temporal firing patterns [516, 518, 523, 590], we targeted this
neuron population for oral cancer classification. At the individual neuron level, the three oral
99
Figure 4.2: Representative images of cell cultures over days. Images are shown for a replicate
of the cell culture used for electrophysiological recordings. All four cell lines are shown over four
days. Healthy cells were observed at all four time points (24-, 48-, 72-, and 96-h).
cancer and the non-cancer VOC mixtures elicited distinct spiking responses over the odor
presentation window. Raw voltage traces of representative extracellular neural recordings
showed clear differentiation between the oral cancer cells, non-cancer cells, and cell culture
medium. We noted differences in PN spiking responses between the three oral cancer cell lines
(Figure 4.4).
Next, we investigated how total spike counts (over the entire 4 s stimulus window) varied
for each recorded neuron corresponding to different VOC exposures (Figure 4.5). To identify
single neurons, spike-sorting of extracellular multi-channel recordings was performed following
100
a f
b c
d e
Figure 4.3: Cell counting procedure and growth curves. a) Schematic of a T25 cell culture flask.
Prior to conducting electrophysiology experiments, the flask for each cell line was imaged 10
times pseudo-randomly. Example imaging locations shown by yellow rectangles. b) Image from
the flask prior to counting is shown. Black scale bar indicates 200 μm. c) An image from a different
location within the same flask is shown. d, e) The same images from b, c are shown post counting
with all live cells marked by a blue dot using FIJI/ImageJ. The mean of the 10 images were taken
to determine the total cell count of each flask. f) Initially, cells were seeded at 1 x 106 cells per
flask at 0 h. Cells were counted from each flask at 24-hour intervals after seeding (24-, 48-, 72-,
and 96-h). At each time point, the total cell count of each flask was averaged over 8seven
replicates. Error bars are S.E.M from the seven replicates. No significant difference in cell counts
were observed between 24-h and 48-, 72-, or 96-h (P < 0.05, d.f. = 6, 16, one-way ANOVA with
Bonferroni correction).
101
Figure 4.4: Headspace from cell lines and control odors evoke unique spatiotemporal patterns
in multiple neurons. VOC-evoked raw neural voltage responses of a recording location are shown
for the three oral cancer cell lines, the non-cancer cell line, the cell culture medium, and two
control VOCs (Hexanal and Undecane). The light grey box indicates the 4 s stimulus presentation
window. Two different recording locations are shown. Position 2 had multiple PNs, which
resulted in different spike amplitudes in the multiunit voltage trace.
previously published methods [591]. Then, we used a simple metric of VOC-evoked time-
averaged and trial-averaged spike counts of individual PNs for each stimulation condition.
Individual PN spike counts were summed over the 4 s stimulus presentation window and
averaged across trials (n = 5 trials) to quantify these changes. Next, we compared the average
spike count of each PN across two stimulus conditions. For example, PN spike counts
corresponding to each oral cancer cell line were compared to the spike counts of the same set of
PNs elicited by the culture medium VOC composition. When all recorded PNs were analyzed,
several PNs showed significant changes in spike counts across two stimulus conditions (P < 0.05,
d.f. = 4, 28, one-way ANOVA with Bonferroni correction). These results demonstrated that there
were differences in individual PN spike counts elicited by cancer vs. non-cancer vs. control VOCs.
Notice that this analysis only compared total PN spike counts corresponding to different stimuli,
102
Figure 4.5: Neurons demonstrate stimulus-specific spiking rates. VOC-evoked total spike counts
(over 4 s) of individual PNs are compared across two stimulus conditions. For each PN, the trial-
averaged total spike count is plotted with the error bars representing S.E.M. of the trial-wise
variations for two stimulus conditions. All comparisons were made with the cell culture medium
evoked spike counts (plotted along the X-axis). All 194 recorded PNs are plotted in every scatter
plot. Individual PNs were identified after spike-sorting of the extracellular recordings. PNs that
responded significantly higher (or lower) to the stimulus VOCs compared to the cell culture
medium VOCs were plotted in red (or blue), respectively (P < 0.05, d.f. = 4, 28, one-way ANOVA
with Bonferroni correction). PNs that did not show significant differences in total spike counts
across two conditions were plotted in grey.
but differences in temporal firing motifs of individual PNs as seen in Figure 4.4 were not reflected
in this analysis.
Oral Cancer Classification via Multi-Dimensional Neural Signal Analysis
To incorporate the temporal spiking characteristics, we analyzed the spatiotemporal PN
responses elicited by the three oral cancer cell lines, the non-cancer cell line, and the cell culture
medium. To generate ‘spatial’ (neuronal identity) – ‘temporal’ (spiking dynamics) response
103
vectors of the entire PN population, trial-averaged firing rates of each neuron were binned into
50 ms non-overlapping time windows. Individual neuron responses were temporally aligned
following stimulus onset. For this analysis, we combined spiking responses from all recorded PNs
Figure 4.6: Cancer vs. non-cancer VOCs are distinguished by spatiotemporal PN responses.
Schematic representation of the spatiotemporal PN response analysis. This analysis contains
spiking responses from all recorded PNs (spatial) and their response dynamics (temporal) over
the 4 s stimulus presentation window. a) Raster plots of all recorded PNs pooled across
experiments are combined for each stimulus (represented by PN1 to PN194). Then the 4 s
duration VOC-evoked spike counts are divided into 50 ms non-overlapping time bins (total 80
time bins for 4 s). Two different stimuli are used for illustration (Ca9-22 and HaCaT). Notice that
the same PNs are recorded for all stimuli. b) A neuron number (n = 194) × time (t = 80) matrix is
constructed, where each element in the matrix corresponds to the spike count of one neuron for
a single time bin (denoted by xitj or yitj). c) The high dimensional population response time-series
is dimensionally reduced using PCA and data corresponding to the first three principal
components are kept. d) The 4 s time-series data is plotted along the three principal component
axes. Each point is connected temporally with the next time point to generate individual VOC-
evoked neural trajectories that take into account both temporal and spatial motifs of the
recorded PN ensemble. Notice that the two PN trajectories corresponding to the Ca9-22 and
HaCaT track along different manifolds in the principal component space. Angular separation
between the two neural trajectories signifies the distinction between the two VOCs. The
percentage of variance captured along the first three principal components is plotted along the
axes. Because of the large number of recorded PNs and their complex response dynamics, the
total variance captured along the first three principal components is low. Therefore, PCA-based
neural trajectories are only used for qualitative comparisons.
104
over multiple days of cell culture. This resulted in a high dimensional population neuron
response, which was represented by an n x m matrix (Figure 4.6a, n = 194 PNs; m = 80 time bins
with 50 ms bin size over 4 s of odor presentation). Next, all recorded PN responses corresponding
to each stimulus were concatenated to generate the population PN time-series data for the
stimulus panel corresponding to Ca9-22, HSC-3, SAS, HaCaT, and the cell culture medium.
To visualize these cell culture VOC-evoked spatiotemporal neural responses, we projected
the high dimensional data onto three dimensions using a linear principal component analysis
(PCA, Figure 4.6d, see Chapter 2: Methodology- Signal Analysis). The points in the three-
dimensional PCA subspace were connected in a temporal order to generate stimulus-specific
neural response trajectories. We observed that each VOC profile generated a closed loop neural
trajectory, which evolved in a unique direction. A long line of work in insect olfaction has
established that the unique direction of the population PN trajectories are specific to odor
identity and intensity [516, 590, 592]. Our previous work demonstrated that larger angular
distances between PN trajectories signify better separability between two odorants [590, 592].
Therefore, unique neural trajectories corresponding to individual VOC mixtures indicate that oral
cancer VOC profiles are distinct from the non-cancer cell line. Moreover, we observed
distinctions among the neural trajectories evoked by the three oral cancer cell lines (Figure 4.7),
which signify that differences between various oral cancers can be identified by this approach as
well.
105
Figure 4.7: Neural responses follow unique temporal trajectories in PCA space. Population PN
trajectory plots are shown after dimensionality reduction using PCA for the three cancer cell lines
(SAS, Ca9-22, and HSC-3), the non-cancer cell line (HaCaT), and the control cell culture medium.
For each stimulus, PN population trajectory is plotted for 0–4 s of VOC exposure. Numbers along
the neural trajectories indicate time in seconds from the stimulus onset. Total number of PNs
used in this analysis is n = 194, which was computed by pooling neurons across all timepoints and
replicates of the cell cultures.
To determine the separation between the cell line specific neural response clusters, we
performed linear discriminant analysis (LDA) on the population PN time-series data (Figure 4.8).
Similar to the PCA analysis, we used the population PN time-series dataset and plotted the VOC-
evoked PN responses in a three-dimensional LDA subspace. This linear dimensionality reduction
technique maximized the neural response cluster separation between stimuli. We observed
distinct clustering of PN responses corresponding to all five stimuli, indicating that a linear
classifier in a three-dimensional LDA space is sufficient to classify cancer vs. non-cancer
successfully based on their corresponding VOC profiles.
106
Figure 4.8: Linear discriminant analysis can elicit clear group distinctions across spatiotemporal
neural responses. Spatiotemporal PN responses (n = 194) for the 4 s stimulus presentation
window (50 ms bin size, total 80 points for each odor) are visualized after dimensionality
reduction using LDA (see Chapter 2: Methodology- Signal Analysis). LDA minimizes within-class
variance and maximizes the variance between classes. Numbers along the axes indicate the
variance captured along that dimension. Distinct clustering of neural responses corresponding to
different VOC profiles indicates that the cell culture VOCs (cancer vs. non-cancer) can be
segregated based on the neural response they elicit.
To get a quantitative estimate of the classification performance, we performed a leave-
one-trial-out cross validation analysis of the PN time-series data (see Chapter 2: Methodology-
Signal Analysis, Figure 4.9). This analysis was performed on the high dimensional dataset (n = 194
PNs, m = 80 time bins) without any dimensionality reduction. During the odor presentation
window, PN spikes are accompanied by a global 20 Hz local field oscillation generated by
synchronized sub-threshold activities of inhibitory local neurons present in the locust antennal
lobe [593-595]. It has been proposed that the Kenyon cell population, which receives direct
inputs from PNs, integrate PN responses within each of these 50 ms time windows generated by
the 20 Hz oscillation. It has been shown that PN responses are most discriminatory within this 50
107
Figure 4.9: Neural responses are well separated in high-dimensional space. Classifications of
VOC-evoked population PN responses without any dimensionality reduction are analyzed by
leave-one-trial-out cross validation analysis (see Chapter 2: Methodology- Signal Analysis). These
quantitative classification results are summarized by a confusion matrix. Each column and row
correspond to the target stimulus and the predicted class, respectively. Here, each 50 ms time
bin of the testing trial is classified as one of the 5 target VOCs based on the minimum Euclidian
distance. The high values along the diagonal of the confusion matrix indicate that most of the
predicted responses match the target labels. This result signifies that information contained
within the 50 ms time bins of the VOC-evoked neural response is sufficient to classify oral cancer
vs. non-cancer and to distinguish different oral cancers from each other. (e) Similar analysis as
shown in panel d except we classified the test trial as a whole by taking the mode of the bin-wise
classification for the 4 s long trial (mode of total 80 time bins for each test trial). This trial-wise
classification of VOC profiles shows flawless distinction of all 5 stimuli tested and reveals the
strength of this neural response-based cancer detection approach.
ms time window. Therefore, to take advantage of this biological neural computational scheme,
we divided the stimulus VOC-evoked population PN responses into 50 ms time bins, starting from
the stimulus onset. For this analysis, Euclidian distances of neural response vectors at each time
bin (50 ms duration) were compared between the testing and the training data (total 80
comparisons over 4 s for each test trial), generating a bin-wise classification (Figure 4.9a). This
bin-wise confusion matrix had its highest values along the diagonal, which implied a high rate of
successful detection of all five stimuli (76% average classification success). Next, we plotted a
trial-wise confusion matrix by calculating the mode of the predicted responses for all 80 time bins
(Figure 4.9b). The trial-wise analysis was implemented to assign one stimulus class value for each
test trial of 4 sec duration. This analysis showed 100% classification for all three oral cancer VOC
108
Figure 4.10: Neural responses follow unique trajectories when including entire odor panel in
PCA space. Ensemble neural trajectories over the 4 s stimulus presentation window are shown
for all 7 stimuli after PCA dimensionality reduction. Volatiles from putative cancer biomarkers
(Hexanal and Undecane 1% v/v diluted in mineral oil) elicited PN responses that traced different
manifolds than those from the cell lines and control media. Numbers along trajectories indicate
time in seconds from the stimulus onset. Total number of PNs used in this analysis is n = 194.
Figure 4.11: LDA dimensionality reduction shows clear distinction between entire odor panel.
Population PN responses corresponding to each stimulus plotted in 3-dimensional LDA space
showed separability between response clusters. Total number of PNs used in this analysis is n =
194.
109
Figure 4.12: High dimensional quantitative analysis enables reliable distinction between entire
odor panel. a) Quantitative classification was performed using a leave-one-trial-out cross-
validation methodology to train and test a linear classifier in the high dimensional feature space.
The time bin-wise confusion matrix shows highest values along diagonal for all cases which
indicates successful classification of all 7 stimuli using PN time-series data. Note the high accuracy
for hexanal and undecane, suggestive that they are mapped to different regions of the encoding
state space b) Trial-wise confusion matrix is plotted for all stimuli. Total number of PNs used in
this analysis is n = 194.
mixtures among themselves and in comparison with the non-cancer and control VOCs. Similar
dimensionality reduction (Figure 4.10 and 4.11) and confusion matrix (Figure 4.12) analyses were
performed on the dataset while including the two other control odorants. Neural trajectories
corresponding to the two control odorants were significantly different from all cell culture VOCs
Figure 4.13: Non-overlapping train-test datasets produces similar results to leave-one-out
based classification. Cancer vs. non-cancer classification using non-overlapping train-test PN
response datasets. VOC exposure trials 1-2 were used for training template construction while
trials 3-5 were used as the test set. This allowed for completely non-overlapping training and
testing sets. Confusion matrices summarize classification results based on a linear classifier in
high-dimensional feature space (see Chapter 2: Methodology- Signal Analysis). a) Bin-wise and
b) trial-wise confusion matrices are plotted for all 5 VOC mixture classified.
110
and the confusion matrix analysis showed high classification success for all seven VOCs tested.
We also performed confusion matrix analysis using completely non-overlapping training and
testing dataset (Figure 4.13), which showed similar results as the leave-one-trial-out cross
validation analysis. We tested bin-wise and trial-wise classification using multiples of 50 ms bin
size (i.e., 100 ms and 200 ms time bins), which yielded comparable results as obtained using 50
ms duration time bin (Figure 4.14).
Time-Matched Cancer Volatile Detection
We anticipated that emitted VOC compositions corresponding to each cell line would vary over
time due to cell growth and ongoing metabolic processes in a fixed cell culture medium. We
also hypothesized that the neuronal template-based VOC classification approach would be able
to compensate for these variations. To investigate this, the neural data that were previously
combined were split and analyzed at four different time points: 24-, 48-, 72- and 96-h after
seeding. All PNs recorded at a specific time point across multiple repetitions of the cell cultures
were combined to generate the population PN response vector for that time point. For
example, each cell culture was repeated 7 times, and the VOC analysis at the 24-h time point
resulted in a total of 42 PNs. All the cell cultures remained viable over 96-h from initiation,
which was verified by manually counting healthy cells at different time points of the cell
cultures (Figure 4.2).
We began by examining VOC-evoked population PN time-series data at 24-h post seeding.
We noticed that dimensionally reduced neural trajectories evolved in different directions for
different VOC profiles in the PCA space (Figure 4.15a). When we performed the same analysis at
111
Figure 4.14: Spatiotemporal PN response-based VOC mixture classification is robust. Three
different bin sizes (50 ms, 100 ms, and 200 ms) were tested for bin-wise and trialwise
classification. Both a, c, e) bin-wise and b, d, f) trial-wise classification results remained largely
similar across different bin sizes used for the high-dimensional VOC classification analysis.
48-h, 72-h and 96-h time points, we continued to observe distinct cell line specific neural
trajectories, which indicated that all the tested stimuli were distinguishable from each other at
different time points of cell growth (Figure 4.15). This observation demonstrated that cultured
cells started emitting VOCs specific to their identity about 24 hours after seeding and remained
separable over multiple days based on their emitted VOC profiles. Next, we analyzed the neural
cluster separation between the three oral cancer cell lines, the non-cancer cell line, and the
112
Figure 4.15: Unique neural trajectories observed at different time-matched points of growth.
PN response trajectories corresponding to the cancer cell lines, the non-cancer cell line, and the
cell culture medium VOC mixtures are plotted after dimensionality reduction by PCA. In this
analysis, neurons were pooled only across experiments performed at predefined time post
seeding. a) A total of 42 PNs were recorded for different cell cultures at this time point. Notice
that neural trajectories are distinct from each other even when the recorded neuron numbers
are lower and the cultured cells are grown for only 24-h. b-d) Cell culture VOC-evoked neural
trajectory plots are shown for 48-, 72-, and 96-h time points of the cell cultures, respectively. At
each time point, PN trajectories traversed distinct manifolds indicating the distinguishability
between cell cultures over multiple days of growth. Note that different PNs (n) were recorded at
different time points.
control medium at different time points of the cultures using LDA (Figure 4.16). Since the number
of PNs recorded at each time point was low, PN response clusters showed some overlap in the
LDA space. This was also reflected in the time bin-wise confusion matrix classification results
performed in the high-dimensional space (Figure 4.17a-d). However, the trial-wise classification
result yielded 100% classification success for each test trial for all the VOCs at all four time points
113
Figure 4.16: Neural response clusters demonstrate visible distinction between stimuli at time-
matched points of growth. a–d) PN population responses cluster distinctly for different VOCs
after dimensionality reduction by LDA at 24-, 48-, 72-, and 96-h of cell growth. In all cases, LDA
shows separability between VOC-evoked neural response clusters.
Figure 4.17: Stimulus-induced neural responses clearly distinguishable in high-dimensional
space at time-matched points of growth. a-d) Time bin-wise high dimensional confusion matrix
analysis of PN responses by leave-one-trial-out approach at 24-, 48-, 72-, and 96-h post seeding.
The confusion matrices have higher values along the diagonal, which indicates that most of the
test trials time bins are classified correctly. However, the confusion matrices also have non-zero
off-diagonal elements, indicative of some misclassification. e-h) Trial-wise confusion matrices are
shown at 24-, 48-, 72-, and 96-h of cell cultures. Here, each test trial was classified based on the
mode of the bin-wise classification results. This analysis elicits diagonal confusion matrices for all
cases, which indicates clear distinction of oral cancer vs. non-cancer VOC profiles based on
population PN spiking responses.
114
(Figure 4.17e-h). Notice that we generated VOC-specific neural fingerprints at each time-
matched points of growth of the cell cultures and performed leave-one-trial-out cross validation
between the test and the training templates generated at that time point.
These results validated our hypothesis that neural response-based classification between
multiple types of oral cancers cells and a non-cancer cell remained unaffected by the variations
in VOCs caused by evolution of cancer and non-cancer cells in the culture medium.
Rapid Detection and Identification of Complex Cancer Volatile Headspace
We investigated how short of a VOC exposure will result in robust cancer classification. We
hypothesized that a neuron response-based classification approach would be fast and able to
classify different VOCs with a short inter-stimulus interval (~ 1 minute). Based on the fast PN
response dynamics, we anticipated that distinction between cancer VOCs would be achieved
Figure 4.18: Root-mean squared (R.M.S) transformation largely preserved stimulus-specific
spiking dynamics. a) Representative recordings from an individual electrode are shown for all
stimuli after high-pass filtering. The gray box delineates the 4 s stimulus presentation period. b)
R.M.S. transformed data traces of panel a recordings reflect the spiking rate-based response
dynamics, while reducing computational overhead. The gray box delineates the same 4 s stimulus
presentation period as in a.
115
within a few hundred milliseconds of stimulus exposure. To achieve fast analyses of neural
signals, we employed a different metric of neural response, which was obtained by root mean
squared (R.M.S.) filtering of raw neuron voltage responses (Figure 4.18, see Chapter 2:
Methodology- Signal Analysis). Until now, all classification analyses were performed after spike-
sorting of multi-unit extracellular voltage responses obtained from each recording location.
However, this approach eliminated neurons that did not pass the statistical test necessary to be
counted as single units. These lost signals from unresolved neurons could potentially be
important for odor discrimination, therefore, we decided to employ the R.M.S.-based approach
which takes into account the total energy of the signal acquired from each location. This approach
was computationally less expensive, unsupervised and shown to be odor specific in our previous
Figure 4.19: Using R.M.S transformed population PN voltage responses to classify the stimulus
panel. Similar plots as shown in Figure S1, but here, we have used R.M.S transformed PN voltages
to generate, a) VOC evoked ensemble neural trajectories after PCA; b) PN response clusters after
LDA; c) Bin-wise confusion matrix in high dimensional space; and d) Trial-wise confusion matrix
corresponding to all 7 stimuli. Spike-based and R.M.S-filtered PN time-series data both yielded
excellent classification for all VOCs tested.
116
work [523]. Using the R.M.S. filtered population PN voltages, we observed distinct classification
of all 7 VOCs tested (Figure 4.19). These classification results were qualitatively similar to the
results obtained from spike sorted single unit data.
To determine the speed and efficacy of this method, we performed VOC classification
during four different 250 ms time segments of the 4 s stimulus presentation window (Figure
4.20). The rationale behind choosing different time windows follows from the unique odor-
evoked response dynamics of the projection neurons. PNs generally fire strongly with high spiking
rates within the first ~1.5 s of stimulus onset, which is known as the ‘transient state’[516-518,
596]. After about 2 s of stimulus exposure, the population PN firing rate converges to a stable
firing rate, which stays above baseline firing but does not change significantly over the rest of the
odor presentation duration. This is known as the ‘steady state’ response period. It is shown in
our and others’ work that odor-evoked transient PN responses are more discriminatory[516, 523,
590]. Therefore, we expected the cell culture VOCs to display the best separation when the
population PN responses are within the transient state. We observed that odor plumes took
about 0.5 s to elicit spiking responses in PNs. This time corresponded to the delay between the
final olfactometer valve opening and the odor plume hitting the antenna. Therefore, we chose
the analysis time windows for transient PN response period as 0.5 – 0.75 s and 0.75 – 1 s and the
steady state time windows as 2 – 2.25 s and 2.25 – 2.5 s (Figure 4.20).
We performed PCA dimensionality reduction analysis to visualize population neural
trajectories, which showed distinct trajectories at the earliest of the time windows (0.5 to 0.75
117
Figure 4.20: Rapid classification of oral cancer VOC profiles using neural voltage responses. a–
d) High dimensional neuron response vector, where each row represents R.M.S. filtered PN
signals of a recording position (n = 84) and each column represents a 50 ms time bin (total 5 time
bins over 250 ms), was dimensionally reduced using PCA and ensemble neural response
trajectories are plotted (see Chapter 2: Methodology- Signal Analysis). Cancer vs. non-cancer
VOC-evoked neural response trajectories are shown for the stimulus presentation windows of
0.5–0.7 5 s, 0.75–1.0 s, 2.0–2.25 s, and 2.25–2.5 s, respectively. Notice that population
trajectories generated from R.M.S. filtered neural voltages are distinct within just 250 ms of odor
exposures. Two 250 ms time windows are shown during the transient state of the PN response
(0–1.5 s), while two other time windows are chosen during the steady state neural response
period (2 s to the termination of the stimulus). e–h) Confusion matrix analysis of the predicted
vs. target responses are shown for the ensemble neural voltage time-series data for the same
time windows as shown in panel a-d. Note that the confusion matrix analysis is done without any
dimensionality reduction. Transient state time windows of 0.5–0.75 s and 0.75–1.0 s show better
VOC classification compared to the steady state time windows. i–l) The same confusion matrices
are plotted for the trial-wise classification, which results into near perfect classification of VOCs
in 250 ms time windows during transient state. Steady state windows show relatively low trial-
wise classification. m) Pairwise distances between ensemble R.M.S. voltages (from n = 84 PN
recordings) corresponding to five different VOCs are plotted in light grey (total 10 pairwise
118
Figure 4.20 (cont’d)
distances) during the stimulus exposure (4 s). The mean of pairwise distances is plotted in black,
which indicates that there are differences in ensemble PN voltage responses corresponding to
different VOCs and these differences are highest during 0.5–1 s of the transient response period.
Note that in our setup, odorants took about 500 ms to reach the antenna from the opening of
the final olfactometer valve (t = 0) and therefore, the earliest transient state time window that
could be chosen was 0.5–0.75 s.
s). The VOC-evoked neural trajectories remained distinct during both transient and steady state
time epochs (Figure 4.20a-d). Next, we performed the quantitative high dimensional confusion
matrix analysis using leave-one-trial-out methodology. We observed better classification during
transient state time windows compared to the steady state time windows, evident from the
higher value of diagonal elements in the confusion matrix shown in Figure 4.20e, f in comparison
to Figure 4.20g, h. Trial-wise classification also showed better predictability during transient state
response periods (0.5 – 0.75 s and 0.75 – 1 s) compared to the steady state segments (2 – 2.25 s
and 2.25 – 2.5 s, Figure 4.20i-l). Finally, when we compared the pairwise R.M.S. response
distances of the PN population elicited by all 5 VOCs, we observed the largest separation was also
during the transient periods. These results demonstrated that once the brain-based recording
and VOC delivery is set up, neuronal population responses can classify VOCs within 250 ms of
stimulus onset.
To verify that a one-minute inter-stimulus interval is sufficient for the VOC classification
and our results are consistent with the PN response dynamics, we employed the R.M.S.-based
classification analysis on the baseline, transient, and steady state epochs of the population PN
response (Figure 4.21). Each analysis epoch was 1.5 s in duration and the 0.5 s delay for the odor
stimulus to reach the antenna was included in the pre-stimulus period. We observed no
119
Figure 4.21: Cancer VOC classification during transient vs. steady state response periods of PN
response. Time bin-wise confusion matrix analysis results shown for three 1.5 s duration time
periods- a) 1 s prior to 0.5 s after final olfactometer valve opening (pre-stimulus period). b) 0.5
to 2 s after final olfactometer valve opening (transient state), and c) 2 to 3.5 s after final
olfactometer valve opening (steady state). The pre-stimulus time period shows no stimulus
specific classification as expected because VOCs had not yet reached the antenna at this time.
The transient state period (0.5 to 2 s) shows the best bin-wise classification of all 5 VOCs. The
steady state period (2 to 3.5 s) also shows high classification success, but the diagonal values are
relatively lower compared to the transient state period. d-f) Trial-wise confusion matrices are
shown for the same time windows as in panel a. g) A population-based peri-stimulus time
histogram (PSTH) plots the change in R.M.S. transformed values of all recording positions (n = 84)
as a function of time. Time labels along the X-axis are relative to the stimulus onset time. A
significant change in R.M.S. values is seen approximately 500 ms after the final valve was opened
(i.e., stimulus onset).
classification in the baseline period (-1 to 0.5 s), but VOC classification was distinct during the
transient (0.5 to 2 s), and steady state (2 to 3.5 s) periods. Overall, VOC-evoked neural responses
120
during the transient period yielded best classification results as expected from the PN response
dynamics [590].
Outlook
While mass spectrometry based techniques have proven essential for volatile chemical
identification for cancer detection [99, 108, 122, 190, 589, 597-616], the desire for portable and
real-time gas sensor has fueled interest in developing electronic nose devices, which mimics
biological principles for odor detection[617, 618]. Electronic noses have increased in popularity
owing to advancements in materials science, nanotechnology, and pattern recognition
algorithms. These devices have demonstrated the ability to distinguish between breathprints of
healthy controls and those afflicted with different types of cancer [216, 263, 269, 602, 619-631].
Although, many of these portable chemical sensors are able to process breath samples in real
time and potentially can be used as point-of-care devices, performance deficiencies due to sensor
drift, humidity and temperature-induced changes significantly complicate implementation as a
breath-based diagnostic. Moreover, electronic noses struggle to attain the high sensitivity
necessary for detecting and differentiating between disease-specific volatiles at the trace-levels
found in breath[617, 618, 632-635]. Despite drawing inspiration from biological olfaction, the
prowess of electronic noses as chemical sensors is minimal relative to their biological
counterpart.
In biological olfaction, natural selection has forced animals to develop highly sensitive
olfactory capabilities while preserving chemical specificity. In the olfactory sensory system, a
target VOC mixture as a whole is encoded by a distinct neuronal response template (or a neuronal
‘fingerprint’ of a VOC), while a different gas mixture is uniquely encoded by a different neuronal
121
fingerprint. It is important to note that biology does not perform component-wise classification
of gas mixtures, but instead achieves optimal separation between the VOC-evoked neural
fingerprints. Many implementations of biological olfaction for disease detection has relied on
behavioral readouts, which require extensive training and significantly limit the response
resolution. Other attempts at creating medical bio-diagnostics have focused solely on the first
order sensors used in animals for chemical recognition and signal transduction[636-646]. This
biohybrid approach of integrating sensory neurons into signal processing platforms has proven
challenging and devices often lack chemical discriminability and long-term performance[618].
Here, we took a forward engineering approach by ‘hijacking’ an insect brain to detect oral
cancers from their VOC signatures. We combined in vivo, multi-electrode, population neuronal
recordings with a multi-channel micro-amplifier, high speed data acquisition, and biological
neural computations to achieve cancer detection. This approach is fundamentally different from
current gas sensing devices and animal behavior-based disease detection as it uses a fully
functional biological chemosensory array (antennae) and olfactory neural circuits as a gas sensor,
and neuronal ‘fingerprints’ of cancer VOC profiles as decoding schemes. This in vivo neural
recording technique can be portable as shown in previous work [523]. We envision this study as
the first step in ‘sniffing out cancer by neurons’ research.
Our brain-based cancer detection method is an untargeted sensing approach for which
we do not need to know the exact chemical composition of the cancer VOC mixture, which is
essential for component-wise classification. Therefore, our method does not identify specific
VOC biomarkers. However, this is also a key strength of biology-based VOC sensing. Biological
olfactory systems detect the entire gas mixture as a single entity (e.g., coffee, banana, specific
122
cancer cell VOC mixture) and encode this information in the brain as a spatiotemporal neural
pattern. By identifying these neural patterns, we can create a template associated with each VOC
Figure 4.22: Principles of neuronal response-based noninvasive cancer detection. a) This
approach employs VOC mixture-evoked neural response templates for distinction between
cancer and healthy samples. During the training phase, target (e.g., breath samples of oral cancer
patients) and control (e.g., healthy human breath samples) VOC mixtures will be exposed to the
locust antenna and in vivo neural recordings will be obtained from multiple projection neurons
in the antennal lobe, simultaneously. The entire in vivo electrophysiological setup will be placed
inside a closed Faraday chamber with inlet and outlet port for VOC delivery and removal. Trained
VOC-evoked population neural responses will be used to construct optimally separated heathy
vs. cancer clusters in the neural space as illustrated in our analysis. Our results indicate that ~40
recorded PNs is sufficient for classification of multiple oral cancer cell lines from healthy controls.
Notice that the training/calibration will be performed for each brain-based sensor, where the
separation between target VOCs will be maximized by optimal placement of the microelectrode
array in the antennal lobe. b) During testing phase, unknown VOC samples (e.g., breath sample
of an early-stage oral cancer patient) will be presented to the antennae and neural responses will
be obtained from the same set of neurons. In our study, we have used the minimum Euclidian
distance between the unknown sample and the healthy vs. oral cancer neural clusters as the
classification metric. However, other distance metrics can be used to classify unknown VOCs.
Since the PN responses reach near baseline within 2 s of odor onset and we have demonstrated
that reliable classification can be performed for 1 min inter stimulus interval, this technique can
work as a high throughput cancer screening device.
123
mixture. Then, an unknown VOC mixture can be tested by comparing it to the known templates
Figure 4.22). In this study, training neuronal templates are generated for the VOC profile of each
cell line and template-matching analyses are done between the training and testing templates to
determine the test cell line identities. By harnessing the gas sensing power of the entire
repertoire of the locust olfactory sensors, which are cross-selective and extremely sensitive, we
can generate distinct neural response templates for different oral cancer cells. Moreover, this
biology-based non-specific detection technology can be generalized to detect other cancers in
the future.
Acknowledgments
A.F. was supported by D.S. The author thanks M. Parnas for assisting in cell culture flask
modification, electrophysiological data collection, spike sorting analysis, and cell counting. E.
Hoque Apu provided support in cell culture seeding and maintenance. E. Cox assisted in locust
colony husbandry and cell counting. N Lefevre and S. Miller assisted in cell counting.
124
CHAPTER 5 | DISCRIMINATING CANCER BIOMARKERS USING HONEYBEE NEURAL
RESPONSES
Application-Specific Chemical Sensors
Chemical gas sensing is a highly interdisciplinary scientific field, drawing upon principles from
molecular chemistry, nanomaterial fabrication, and signal analysis to name but a few. One reason
why it is such an intricate process is due to the millions of chemical combinations in existence.
These molecules can be combined in highly heterogeneous matrices, forming exceptionally
complex stimuli. The resultant stimulus state space would simply be infeasible to cover with a
labeled line approach, in which individual sensors were responsible for encoding specific stimuli.
Fortunately, stimuli are often constructed in stereotypical manners, relying heavily on relatively
basic yet highly efficient chemical configurations. This chemical redundancy has prompted
sensors, such as electronic noses and biological olfactory neurons (ORNs), to implement a
combinatorial pattern recognition scheme. In this case, sensors can be tuned to highly conserved
features, which can then be combined into ensembles to efficiently encode a large number of
stimuli[647]. By incorporating an assortment of highly selective as well as broad range sensors,
this approach can optimize chemical sensitivity, selectivity, and generalizability based on a finite
stimulus space[371, 648]. This can be thought of in comparison to our sense of hearing. Within
the ear, inner hair cells are arranged in a tonotopic pattern with adjacent cells being maximally
responsive to sounds of near similar frequencies. Importantly, the response curves for a number
of successive hair cells overlap, enabling a much finer resolution of tone discrimination[649]. The
overlapping sensitivity between different sensors is a critical component for systems
incorporating combinatorial coding information processing schemes.
125
Olfactory receptor neurons (ORNs) and electronic nose sensors display variability in their
tuning curves. That is while one sensor may be highly selective and display high affinity for a
particular chemical, others are responsive to a broad range. In electronic nose sensor
development, the guiding force is based on an iterative process of fabricating or combining new
materials and testing their response characteristics to a particular stimulus. While great strides
in electronic nose technology have been made since its inception 40 years ago, state-of-the-art
devices struggle to achieve sufficient detection limits for the wide range of disease-specific
volatiles found in exhaled breath. While improvements in sensor material technology and signal
processing algorithms will improve the gas sensing abilities of electronic noses, the performance
of current systems lags behind that of biological olfaction. In biology, the force dictating the
response characteristics of ORNs is evolution via natural selection. It reasons that ORNs providing
a selective advantage for an organism within its immediate environment, are much more likely
to be maintained through successive generations. For example, humans are highly responsive to
odors in blood, information critical for assessing injury or danger, while unresponsive to carbon
monoxide, as prior to the advent of fire, it provided no selective advantage[650, 651]. As such,
organisms display species-specific olfactory receptor genes and differentially tuned ORNs based
on unique environmental pressures.
Previous research on insect olfaction has demonstrated that locusts respond to a broad
range of odors, from those that hold biological significance to those that have had no influence
in their evolution. Moreover, even these novel, non-biologically relevant stimuli were mapped to
different regions of their chemical encoding state space, allowing for efficient signal
discriminability. Our previous work has demonstrated that locusts can not only identify unique
126
cancer biomarkers, but their olfactory system is powerful enough to differentiate between the
complex volatile headspaces of cancer and non-cancer cells. Like all sensors, however, locust
olfaction has a limited scope across a subset of possible odor space. It is improbable that the
locust olfactory system, which has developed to function optimally in a particular ecological
niche, is the best sensor for classifying disease from breath samples.
Honeybee Olfaction
A significant amount of research has been conducted using honeybees, due to their impressive
learning abilities, gregarious nature, and critical role in the environment[652]. Like other insects,
honeybees rely strongly on their olfactory abilities, rapidly identifying pertinent volatiles,
localizing plumes and occasionally using them for long-distance flights[653]. As a result of wide
interest in honeybee neurobiology and behavior, multiple standard brain atlases have been
constructed, aiding in understanding network topologies and functional processing schemes
(Figure 5.1) [654-657]. A variety of studies incorporating genomic, behavioral,
electrophysiological, and calcium-based imaging data, have elucidated key anatomical and
functional features specific to honeybees that are integral for their highly acute chemical sensing
abilities[498, 658, 659]. Preliminary research has indicated that honeybees can be conditioned
to engage in stereotypical behavioral responses to highly specific stimuli. Importantly, like other
biological olfactory systems, they analyze complex patterns of chemoreceptor activations to
recognize relative concentrations of chemical constituents and identify unique odorants[499]. A
deeper look into their anatomical and functional neurobiology can help to unveil some of the
mechanisms used to perform this impressive feat.
127
Figure 5.1 | Surface model of the honeybee standard brain. Olfactory receptor neurons
converge onto glomeruli in the antennal lobe. Here, two distinct populations of glomeruli are
shown, T1 and T3. These neuronal populations project axons along particular paths to higher-
level processing areas such as the median calyx (MCA) and lateral calyx (LCA) of the mushroom
body (MB). Scale: 250 mm. Reproduced from [657].
Honeybees contain three types of olfactory-specific sensilla: placoid (poreplate), trichoid
(hair-like), and basiconic (peg-shaped)[660]. The placoid sensilla contain between five and 35
ORN dendrites[323, 661]. Hydrophobic volatiles are bound by odorant binding proteins and
transported to the appropriate ORN. Interestingly, honeybees have the smallest set of odorant
binding proteins (21) of all studied insects[662] but a rather expansive set of ORNs (163)[663].
While the activity of one ORN can alter that of its neighbors due to the shared lymphatic ionic
environment, direct ligand binding competition is rare as neurons with similar tuning curves are
likely to be found in different sensilla[664]. A total of 60,000 ORNs extend axonal processes into
the antennal lobe, where a portion splits into four different tracts and innervates the 165
128
glomeruli[665]. Like most other insects, ORN axons of a particular type innervate a single
glomerulus[666]. Within each glomerulus, a stereotypical topography is observed: ORN axons
innervate the cortex and PNs innervate the core[654, 666]. The honeybee antennal lobe consists
of 800 projection neurons and 4000 local neurons. This is in striking contrast to the locust
antennal lobe, which displays 830 projection neurons and 300 local neurons. Differences in total
neuronal populations suggest that these insects use highly disparate processing schemes in order
to effectively encode olfactory stimuli. Moreover, the local neurons in locusts issue graded
dendrodendritic potentials, whereas those in honeybees fire sparse action potentials. The
inhibitory local networks in both species play a vital role in effectively encoding different
odorants[364]. Projection neurons transmit conditioned signals to higher-level brain regions,
such as the mushroom body characteristic of signals divergence and sparse neuronal
activity[365]. Each glomerulus can be thought of as encoding for a particular stimulus related
feature. On average, a single glomerulus contains axons from 400 ORNs, 1000 local neurons and
a mere five projection neurons[323, 667]. This crude breakdown indicates massive convergence
of chemical signals onto a small number of extrinsic neurons, emphasizing the critical nature of
temporal dynamics for maximizing discriminability.
Honeybee-Based Cancer Biomarker Differentiation
In our current study, we sought to determine the efficacy of the honeybee olfactory system in
detecting and discriminating between key cancer biomarkers found in exhaled breath.
Microelectrodes were used to record neuronal action potentials from the antennal lobe. We
included a panel of 10 odorants, identical to those used in our previous research intended to use
the locust olfactory system to identify individual cancer biomarkers. Our stimulus panel consisted
129
of five odors affiliated with breast cancer (hexanal, nonanal, pentanal, trichloroethylene, and
undecane), five odors affiliated with lung cancer (decane, methylheptane, pentamethylheptane,
propylbenzene, and undecane) and one stimulus (paraffin oil) as a negative control. Note that
some biomarkers, such as undecane, are observed to be elevated in a number of different
cancers owing to common underlying pathophysiological adaptations. All odorants were diluted
in paraffin oil (1% v/v) and stored in 20 mL glass vials with 1/32” diameter
Figure 5.2 | Peri-stimulus voltage trace. Honeybee antennal lobe neurons were initially
processed using an RMS transform and population average voltages were plotted as a function
of time. Here, different stimuli seem to elicit unique response patterns corresponding to
stimulus-specific spatiotemporal encoding of stimuli.
130
polytetrafluoroethylene (PTFE) tubing serving as inlet and outlet lines. For additional information
on odorant preparation, see section Chapter 2: Methodology- Odor Vials. All stimulus
presentation periods lasted for four seconds. For additional information on odorant preparation,
see section Chapter 2: Methodology- Odor Presentation.
Similar to the encoding state space of enoses, the responses of individual electrodes or
neurons can be conceptualized as unique orthogonally oriented dimensions in an encoding
hyperspace[203]. Given current limitations in microelectrode design and fabrication, we pooled
the total number of electrode positions across 39 different individuals. This technique, well-
established for other insect models, is based on a breadth of data demonstrating that response
patterns remain consistent between species-specific individuals[498-500].
Owing to the positive results gained from locust research, we processed data with our
RMS technique. Population-wide neural responses demonstrated honeybees successfully
detected the presented odorants (Figure 5.2). We observed unique response dynamics with a
large transient response for the majority of the presented stimuli. This could potentially allow for
stimulus discrimination based on neural responses alone. We performed similar dimensionality
reduction techniques as we had done with our locust dataset. Initially, we plotted neural
trajectories after processing the data according to principal component analysis (Figure 5.3a).
This again showed unique neural trajectories for each individual odorant. Data mapped to and
plotted in linear discriminant analysis subspace demonstrated disparate stimulus-specific neural
cluster formation (Figure 5.3b). We performed quantitative analysis using our leave-one-trial-out
technique. Bin-wise predictions and trial-wise predictions were plotted in a confusion matrix for
the four second period following stimulus onset. This constituted the entirety of the odor
131
Figure 5.3 | Dimensionally reduced population-wide response dynamics. a) Principal
component analysis allows for visualization of neural response trajectories according to the
degree of variance they express. Highly unique trajectories indicate that the response state
corresponding to each stimulus at each point in time differ dramatically from one another. b)
Stimulus-specific clusters begin to emerge from linear discriminant analysis processing. This
suggests that an underlying structure exists for encoding and processing this data in the
honeybee brain and these responses may be linearly separable with enough recording locations.
presentation period, including the initial transient response and subsequent steady state. Bin-
wise predictions showed adequate classifier performance (Figure 5.4a). Trial-wise performance
was near perfect, with only two out of 50 trials predicted incorrectly (Figure 5.4b).
Figure 5.4 | Classification of cancer biomarkers. We used a leave-one-trial-out method for
training and testing a linear classifier in the original 39-dimensional neural encoding state space.
a) Results from bin-wise classification show moderate levels of classification accuracy using the
Manhattan norm. b) Trial-wise based classification generates near perfect results. A total
accuracy of 96% was attained using the Manhattan norm as a distance metric.
132
Outlook
It is important in gas sensing applications to consider the particular application as well as the
strengths and weakness of different sensor technologies. This iterative process is relatively
straightforward for manmade sensors, but a nearly limitless number of confounding factors can
threaten readout integrity. It is also immensely time-consuming as application-specific devices
must be fabricated, each tested within an assortment of different environmental conditions .
Biological olfaction has evolved to function over a wide range of environments. It is likely less
affected by the high humidity levels found in exhaled breath and is robust to minor alterations in
background noise or interferants. This ability stems from the highly intricate processing schemes
that incorporate spatial and temporal aspects for enhanced encoding capabilities. This imbues
biosensors as highly attractive models for processing complex gas mixtures. In this case,
biosensor tuning is likely based on evolutionary pressures forcing organisms into particular
ecological niches. Having demonstrated that the locust can effectively detect cancer biomarkers,
we sought to test the same capabilities in the honeybee, which exhibit disparate sensory
processing capabilities. We found that honeybees can, in fact, differentiate between stimuli with
a high degree of accuracy. While tempting to compare the classification accuracies between
locusts and honeybees as an indicator for olfactory performance, it should be noted that different
technologies for interfacing with the neural system were utilized owing to alternate
morphological features of the two species. To our knowledge, this study represents the first
extracellular recordings from the honeybee antennal lobe and, hopefully, paves the way for
future experiments using honeybees as a non-invasive cancer diagnostic.
133
Acknowledgments
A.F. was supported by D.S. The author thanks E. Cox and S. Sanchez for performing odor vial
construction and electectrophysiological recordings. E. Cox, S. Sanchez, and M. Parnas also
provided support in honeybee colony maintenance and husbandry.
134
CHAPTER 6 | ONGOING WORK AND FUTURE DIRECTIONS
Introduction
Bio-hybrid sensors offer fundamental advantages over current state-of-the-art manmade sensors
for the purpose of disease detection. Gas chromatography-mass spectrometry and ion mobility
spectrometry offer excellent component wise separation abilities. These analytical approaches
have provided insight into underlying metabolic processes by identifying analytes tied to specific
biological pathways. However, the goal of a medical diagnostic is to identify the presence (or
absence) of an underlying disease. While individual volatiles can be indicative of a few different
pathological conditions, breath profiles are combinations of hundreds to potentially thousands
of trace-level compounds. Subtle variations in a broad range of these chemicals are often much
more informative for disease diagnoses, especially when considering complex pathologies.
Importantly, the production of intermediary metabolites can alter subsequent biochemical
reactions, eliciting further changes in metabolites. As such, diseases often have broad scale
effects dependent on numerous genetic as well as environmental variables. These component-
wise breath analysis methods as medical diagnostics require extensive signal processing to
identify patterns particular to a disease state. In order to bypass the limitations of such systems,
electronic noses are based on a functional combinatorial coding scheme. The incorporation of
broadly selective, cross-reactive sensors prevents the identification and quantification of
individual analytes. Instead, these devices aim to classify entire gaseous mixtures as points in a
high-dimensional space. Although inspired by biological olfaction, the performance of electronic
noses pale in comparison. They lack the broad response characteristics seen in biological
olfactory systems and fail to achieve sensitivity levels for the detection of many trace-level
135
volatiles present in exhaled breath. Instead of attempting to reverse-engineer gas-based sensors,
our approach harnesses the power of biological olfaction directly by incorporating complete
biological chemosensory arrays involving diverse olfactory receptors (antennae), biological signal
transduction, and neural computations all in one device. We have demonstrated that multiple
different insect biosensors are able to reliably detect and differentiate between cancer
biomarkers. Moreover, the locust olfactory system is capable of differentially encoding
heterogeneous mixtures such as the volatile headspace of cancer cells. Our proof-of-concept
testing indicates that insect biosensors have tremendous potential as medical diagnostics.
Current Limitations
All insect brain surgeries were performed manually. For locusts, each surgery took one to one
and a half hours depending on the experimenter’s surgical experience and expertise. Success rate
of electrode insertion was moderate (~80% per surgery). In future iterations, these limitations
can be mitigated by incorporating robotic surgeons to automate the surgery and electrode
insertion processes[668-670]. In our lab, the supply of insects was relatively abundant and
inexpensive. For large scale operations, significant increases in colony sizes would be necessary,
requiring some trained operative personnel.
One significant concern regarding brain-machine interface technologies is the potential
damage to neurons and glial cells. While for acute experiments this has limited impact, the
induced damage can initiate an inflammatory response and significantly affect signal-to-noise
ratios in chronic models[671]. Based on our current neural recording setup, the number of PNs
recorded per experiment was low (on average ~3 PNs/recordings). Therefore, PN responses were
pooled across multiple recordings which is standard in neuroscience studies[523, 590, 596] but
136
not optimal for one-shot cancer VOC analysis. In the future, high density multi-electrode arrays
and signal processing circuitry can be employed to record large amount of PNs simultaneously,
thereby eliminating the need for multiple experiments and achieving one-shot classification [670,
672-674].
In the locust antennal lobe, projection neurons are the only neurons to exhibit a spiking
response. Thus, we each class of spike sorted data belonged to a putative projection neuron. In
honeybees, however, both local neurons and projection neurons exhibit spike-based signaling.
We could not positively determine whether our spike sorting algorithm was identifying a local or
projection neuron, though the activity of both might be beneficial for stimulus classification.
Waveform analysis algorithms may be effective in discriminating between the functional
characteristics and separating out activity from these distinct populations.
One significant concern of manmade sensors as medical diagnostics is their limited
sensitivity. While biological olfactory systems, such as canines, have shown to exhibit very low
detection thresholds, our current study did not investigate whether these insect biosensors could
detect and differentiate between volatiles or volatile mixtures at the concentrations typically
found in exhaled breath. The selection of cell culture stimuli in our locust study enabled a highly
controlled environment limiting the presence of confounding variables. However, in vitro cell
cultures do not fully reflect the in vivo tumor microenvironment and cellular matrix, and in vivo
VOC profiles may differ from the VOC profiles observed in in vitro cultures for different cancer
types [146, 147, 675, 676]. Therefore, further studies involving in vivo cancer VOC analysis will be
necessary to validate this brain-based sensing technology for cancer detection.
137
Multi-Electrode Array Modifications
Sensor response characteristics are vital for determining a system’s performance in a given task.
By tapping into the neural signals from insects, we eliminate the difficulty in designing and testing
effective sensors for a broad range of chemicals. Yet, we still need optimal technologies to read
the electrical signals that have been transduced by olfactory receptor neurons and processed by
the dense neural network of the antennal lobe. State-of-the-art multi-electrode array fabrication
technologies are generally geared towards device implantation in humans to restore normative
sensory functioning. As such, the dimensions of these arrays are not suitable for insect-based
neural recordings. We explored different avenues for the creation of novel high-density electrode
arrays that eliminate the need to pool data across multiple insects. The antennal lobes of locusts
and honeybees are ~500µm and ~300µm in diameter, respectively. For our locust recordings, we
used micromachined probes, however, microwire arrays minimize the amount of displaced tissue
Figure 6.1 | Multi-tetrode device. A single guide tube contains five different tetrodes, consisting
of four channels each. While this technique could be beneficial for recording from a large number
of neurons, it lacks reproducibility as tetrode bundles are not stabilized within the guide tube.
138
Figure 6.2 | Multi-electrode array size comparisons. Left to right: Dual shank Neuronexus probe
with two tetrodes per shank, 4-, 8- and 16-channel twisted wire tetrodes. As the number of
channels increases for the twisted wire tetrodes so too does the device footprint. For insertion
points with no space between individual wires, device implantation was not viable and a
significant amount of neural damage occurred.
Figure 6.3 | Dual guide tube single wire electrodes. Individual 12.7 µm diameter wires were fed
through micro-capillary columns with an inner diameter of 50 µm. The hole—wire size mismatch
permitting significant maneuverability within the tubes and the relatively large tube thickness
limited using a large number of tubes for a highly dense array. Line-to-line width in the above
images is 200 µm.
139
and are often characterized by much smaller device profiles. As such, twisted wire tetrodes were
used for honeybee recordings to minimize the extent of insertion damage. For locusts, the larger
working area permitted multiple tetrodes to be paired within the same device (Figure 6.1).
However, inter-tetrode pitch was highly variable and twisting of the wires created a solid mass
at the insertion point (Figure 6.2), issuing substantial redundancy among a tetrode’s channels.
Initial attempts to reduce the electrode footprint and maximize information content were
attempted by inserting individual electrode wires into adjacent tubes to enable sufficient inter-
electrode separation (Figure 6.3). Yet, the resulting gap between electrodes was too large to
enable the construction of high-density arrays. Next, we attempted to splay electrode wires at
the ends of a spun tetrode (Figure 6.4). The splayed channels achieved appropriate inter-
electrode distances but, similar to the variable pitch between tetrodes observed in our multi-
tetrode devices, the pitch between individual wires varied significantly. This technique not only
produced inconsistent channel spacing but was difficult to scale up to polytrodes containing the
high number of channel counts intended. Instead a jig template was designed and printed using
a Stratasys J750 3D printer (Figure 6.5). Test pieces containing 400 µm, 300 µm, 200 µm, and 100
Figure 6.4 | Flared multi-electrode array. Tetrodes were fabricated as seen in Chapter 2:
Methodology- Tetrode fabrication. Once finished construction, tetrode tips were splayed to form
close, yet identifiably separate channels. These devices proved viable but lacked inter-device
reliability and proved difficult to scale up to higher channel counts. Line-to-line distance is 200
µm.
140
Figure 6.5 | 3D-printed multi-electrode template. Template piece for testing Stratasys J750
resolution and minimal feature size capabilities. Printer successfully printed 400 µm and 300 µm
holes but failed to print 200 µm and 100 µm holes. While the larger holes were visible, the
channel did not extend the entire way through the piece.
µm diameter holes were initially printed to determine the maximal feature resolution of the
printer (Figure 6.5a). Holes exceeding 150 µm failed to be printed (Figure 6.5b). This test showed
that the resolution of the printer was inadequate for generating jigs with necessary inter-
Figure 6.6 | Boston Micro Fabrication 3D-printed multi-electrode template. A template piece
for testing the resolution and minimal size feature capabilities of a 3D printer specializing in
micro-scale technology was printed. Hole diameters included 80 µm, 70 µm, 50 µm, 30 µm, 25
µm, 20 µm, 18 µm, 15 µm, 12 µm, and 10 µm. The printer successfully printed most, if not all, of
the holes, though this was difficult to assess for the smaller ones even with the use of a
stereomicroscope. a-d) Template test piece. e) Supplementary 3D printed holding device to
stabilize piece during wire feed through. f, g) Wires successfully fed through the larger holes of
the device, demonstrating excellent printer resolution. h) Custom-made 3D multi-electrode array
just prior to device insertion.
141
electrode pitch distances and the large hole size would induce variability in pitch distance. A new
test piece printed by a company specializing in micro-scale 3D printing applications (Boston Micro
Fabrication). Holes of various diameters were tested, and the printer achieved resolution
adequate for our intended application (Figure 6.6). Handling of the 3D printed jig template was
exceptionally difficult owing to its small profile and extremely light weight. However, some
Figure 6.7 | 3D multi-electrode array. Close up images of an additional 3D multi-electrode array
as seen through a stereomicroscope.
142
a) b)
Figure 6.8 | Alternative multi-electrode array designs. a) Custom-design 3D printable horizontal
jig. b) Micro-capillary column with numerous microscopic hole channels.
electrode wires were able to be fed through individual holes and a layer of two-part epoxy was
placed atop the template, thereby stabilizing electrode positioning. Once cured, the epoxied wire
was removed from the jig template to expose tips of the 3D high-density probe (Figure 6.7).
Alternative design schematics were considered but not pursued (Figure 6.8a). Moreover,
microcapillary columns, occasionally used in gas chromatography-mass spectrometry and ion
mobility spectrometry, could be used to serve as a template substrate (Figure 6.8b).
In addition to 3D geometries, 2D configurations were also explored. One design used
wires aligned in parallel, which resulted in excellent configurations, but the fabrication of such
probes was painstaking and difficult to reproduce (Figure 6.9). Alternatively, wires were drawn
a) b) c)
Figure 6.9 | 2D straight-wire multi-electrode array. a) Test array of two individual wires. b, c)
Test array demonstrating multi-channel capabilities at micron level resolution.
143
a) c) e)
b)
d) f)
Figure 6.10 | 2D angled-wire multi-electrode array. a, b) Test array of two individual wires. c-f)
Array testing multi-channel capabilities at micron level resolution.
at angled geometries crossing over a central point. Cyanoacrylate adhesive was applied to
stabilize the wires adjacent to the point of cutting. However, even with sharp microscissors,
minor hand movements put undue tension on the wires and unintentionally altered relative tip
positions (Figure 6.10). Laser cutting could be applied to mitigate tip position alterations, but this
avenue was not explored.
Figure 6.11 | Dual-lobe multi-tetrode device. Both antennal lobes could be targeted at the same
time, doubling the number of recording locations and potentially putative spike sorted neurons.
144
a) b) c)
Figure 6.12 | Planar multi-electrode array. Microfabricated flexible surface-level electrode
arrays targeting both antennal lobes. a) PCB board facilitating electrode connections. b, c) Planar
array implanted on top of the locust antennal lobe.
In addition to microwire arrays, a planar array device was custom fabricated and tested
for in vivo recordings (Figure 6.12). This flexible device was fitted with multiple surface electrodes
intended to record from the projection neurons residing in the cortex of the locust antennal lobe.
Unfortunately, even with hydrophilic and hydrophobic film applications, the planar array did not
adhere to the antennal lobe surface sufficiently. A future microfabricated iteration consists of a
foldable device with penetrating electrode shanks (Figure 6.13). Probes can contain a variety of
individual electrode channels, however, these have yet to be tested in vivo.
Other electrode modifications could be made to enhance recording capabilities that do
not necessitate increasing channel density. Electrochemical etching of the wire tips could reduce
the electrode tip footprint, thereby minimizing probe insertion damage. This could potentially
increase device longevity and improve signal-to-noise ratios. Alternative electrode wire materials
that have enhanced conductivity could be explored. While for the majority of brain implants, the
stiffness or young’s modulus of a metal ought to be considered, removal of the neural sheath
covering the antennal lobe exposes an extremely soft surface. This requires minimal material
rigidity for penetration, allowing electrode material selection to optimize electrical properties
145
with little mechanical constraints. We electroplated the wire tips with pure gold for all
experiments, but other plating materials, such as carbon nanotubes, that may enhance signal-to-
noise ratios may be considered.
Sensor Calibration
Device longevity and sensor calibration are significant concerns for any chemical sensor,
especially those incorporating biological componentry. Optical imaging techniques benefit from
the topological and functional glomerular conservation observed between individuals within the
same species. This allows for the generalization of odor-specific glomerular response profiles
across different individuals. For biosensors incorporating electrode-based neuronal signals, it is
highly unlikely that electrodes will be positioned at the exact same location in different
individuals. Therefore, sensor calibration procedures ought to be performed for each individual
based on the encoding state space generated by the electrode array’s positioning within the
antennal lobe. This sensor initialization process is effective due to the spatiotemporal
combinatorial coding scheme implemented by the insect olfactory system. Instead of relying on
the identity of individual neurons, distinct neuronal response training templates can be formed
from the entire signal population of putative neurons or recorded electrodes. While some
variation will be observed between the effectiveness of each individual sensor, ensembles of
broadly tuned projection neurons will recognize stimulus-specific features and encode disparate
stimuli in unique manners.
Various experimentally derived parameters can be selected for the application at hand.
For example, during training template formation, a maximum of 10% overlap could be allowed
between population PN responses. Additionally, during the testing phase, a minimum of 80%
146
classification success rate may be chosen. The tuning of these parameters enables altering the
balance between accuracy, precision, and recall, a highly useful ability when implementing the
chemical sensor as a screening tool versus a later stage diagnostic to determine disease
progression.
The limited system lifetime is offset by significant sensor reversibility, which promotes
high sample throughput capabilities. Unlike the substantial processing and recovery times
associated with artificial sensors, the insect olfactory system is capable of rapidly binding,
Figure 6.13 | Excised locust antennae and brain. The brain was able to be extracted but
significant strain was enforced upon the antennal lobe, corrupting incoming electrical signals.
147
sensing, and removing odorant molecules. The real-time processing capabilities of this
biologically based chemical sensor allows for odorants to be delivered on the second- and
minute-timescale. Repeated presentations with such low intervals could serve to substantially
increase stimulus encoding and effective discrimination.
Brain-on-a-Chip
In an effort to move towards highly portable devices, we are actively exploring brain-on-a-chip
technologies. Essentially, the insect body is unnecessary as long as the antennae neural circuitry
remains intact. Efforts to remove the locust antennae and brain as chemical sensors and an
Figure 6.14 | Ex vivo v2. Attempts were made to decapitate the locust and subsequently perform
a brain extraction method. The inserts seen above slid into a secondary surgical reservoir filled
with locust saline to mitigate tissue desiccation. However, a lack of head stability prevented this
technique from being effective.
148
associated processing unit have been undertaken. Preliminary extraction methods were
attempted by performing usual in vivo brain surgery followed by excising the exoskeleton
surrounding the antennae and brain. We were able to keep structures intact, but these methods
induced considerable damage to the antennal lobe and nerve, thus corrupting signal transmission
(Figure 6.13). Other efforts included a decapitation method and subsequent antennae and brain
excision in a secondary 3D printed surgical basin (Figure 6.14). Here, however, the structure of
the basin complicated surgical accessibility and the locust head did not remain stable during
surgery. A third strategy involved designing an entire in vivo surgical platform followed by
antennae—brain extraction with particular focus on preserving the integrity of the antennal
nerve (Figure 6.15). The device proved to be effective during surgical procedures but attempts
to determine whether stable electrophysiological-based recordings can be attained have not yet
been conducted.
Figure 6.15 | Ex vivo v3. A novel design was cadded and 3D printed. The device performed as
expected during the surgical procedure, however brain extraction techniques have not yet been
attempted.
149
In Vivo Diagnostic Validation
Apart from the aforementioned innovations, further research should be conducted incorporating
more realistic models to determine the diagnostic feasibility of our insect-based biosensor. Such
testing can include 3D cell culture models, volatiles from diseased animal models or preclinical
experiments processing breath samples from human patients.
Outlook
Manmade gas sensing technologies are powerful and exhibit a number of positive features.
Component-wise methods are integral technologies for exploring the individual analytes of a
mixture and for identifying chemical concentrations. In particular, gas chromatography-mass
spectrometry is an essential method for laboratory research in a variety of scientific domains.
Electronic noses offer rapid sample processing times and high potential for device miniaturization
and portability. However, no manmade sensor has demonstrated the impressive chemical
sensing capabilities characteristic of biological olfactory systems. We have successfully
demonstrated that both locusts and honeybees can be used to detect and differentiate between
cancer biomarkers. We believe that using biosensors as chemical sensing devices provides a
number of advantages over conventional sensors. While biological olfaction has been lauded for
its impressive sensitivity and broad ranging specificity, the development of biosensors as medical
diagnostics is an interdisciplinary approach that requires expertise in a number of areas. As
related technologies and signal processing algorithms continue to improve, the potential of the
field will follow suit. We hope that this research paves the way for future investigations of novel
biosensors for disease detection and other practical, real-life applications.
150
REFERENCES
1. Davis, C., J. Pleil, and J. Beauchamp, Breathborne biomarkers and the human volatilome.
2020.
2. Domínguez, R., et al., A comprehensive review on lipid oxidation in meat and meat
products. Antioxidants, 2019. 8(10): p. 429.
3. Wilson, A.D. and M. Baietto, Advances in electronic-nose technologies developed for
biomedical applications. Sensors, 2011. 11(1): p. 1105-1176.
4. Pavlou, A. and A. Turner, Sniffing out the truth: clinical diagnosis using the electronic
nose. 2000.
5. Asha, P., et al., IoT enabled environmental toxicology for air pollution monitoring using
AI techniques. Environmental Research, 2022. 205: p. 112574.
6. Fan, W. and J. Almirall, High-efficiency headspace sampling of volatile organic
compounds in explosives using capillary microextraction of volatiles (CMV) coupled to
gas chromatography–mass spectrometry (GC-MS). Analytical and bioanalytical
chemistry, 2014. 406(8): p. 2189-2195.
7. Miekisch, W., J.K. Schubert, and G.F. Noeldge-Schomburg, Diagnostic potential of breath
analysis—focus on volatile organic compounds. Clinica chimica acta, 2004. 347(1-2): p.
25-39.
8. Dent, A.G., T.G. Sutedja, and P.V. Zimmerman, Exhaled breath analysis for lung cancer.
Journal of thoracic disease, 2013. 5(Suppl 5): p. S540.
9. Das, S., S. Pal, and M. Mitra, Significance of exhaled breath test in clinical diagnosis: a
special focus on the detection of diabetes mellitus. Journal of medical and biological
engineering, 2016. 36(5): p. 605-624.
10. Kim, K.-H., S.A. Jahan, and E. Kabir, A review of breath analysis for diagnosis of human
health. TrAC Trends in Analytical Chemistry, 2012. 33: p. 1-8.
11. Porter, R., The greatest benefit to mankind: a medical history of humanity (the Norton
history of science). 1999: WW Norton & Company.
12. Hayden, G.F., Olfactory diagnosis in medicine. Postgraduate medicine, 1980. 67(4): p.
110-116.
13. Fitzgerald, F.T. and L.M. Tierney Jr, The bedside Sherlock Holmes. Western Journal of
Medicine, 1982. 137(2): p. 169.
151
14. Kelly, M., Scent of a patient: an underestimated role in clinical practice? British Journal
of General Practice, 2012. 62(600): p. 378-378.
15. Buszewski, B., et al., Human exhaled air analytics: biomarkers of diseases. Biomedical
chromatography, 2007. 21(6): p. 553-566.
16. Ortega, R., et al., Monitoring ventilation with capnography. The New England journal of
medicine, 2012. 367(19): p. e27-e27.
17. Kodali, B.S., Capnography outside the operating rooms. The Journal of the American
Society of Anesthesiologists, 2013. 118(1): p. 192-201.
18. Haick, H., et al., Assessment, origin, and implementation of breath volatile cancer
markers. Chemical Society Reviews, 2014. 43(5): p. 1423-1449.
19. Nardi-Agmon, I., et al., Exhaled breath analysis for monitoring response to treatment in
advanced lung cancer. Journal of Thoracic Oncology, 2016. 11(6): p. 827-837.
20. Taylor, D., et al., Exhaled nitric oxide measurements: clinical application and
interpretation. Thorax, 2006. 61(9): p. 817-827.
21. Barnes, P.J., et al., Exhaled nitric oxide in pulmonary diseases: a comprehensive review.
Chest, 2010. 138(3): p. 682-692.
22. Dweik, R.A., et al., An official ATS clinical practice guideline: interpretation of exhaled
nitric oxide levels (FENO) for clinical applications. American journal of respiratory and
critical care medicine, 2011. 184(5): p. 602-615.
23. Smith, A.D., et al., Diagnosing asthma: comparisons between exhaled nitric oxide
measurements and conventional tests. American journal of respiratory and critical care
medicine, 2004. 169(4): p. 473-478.
24. Owen, O., et al., Acetone metabolism during diabetic ketoacidosis. Diabetes, 1982.
31(3): p. 242-248.
25. Tassopoulos, C., D. Barnett, and T.R. Fraser, Breath-acetone and blood-sugar
measurements in diabetes. The Lancet, 1969. 293(7609): p. 1282-1286.
26. Wang, Z. and C. Wang, Is breath acetone a biomarker of diabetes? A historical review on
breath acetone measurements. Journal of breath research, 2013. 7(3): p. 037109.
27. Broza, Y.Y. and H. Haick, Nanomaterial-based sensors for detection of disease by volatile
organic compounds. Nanomedicine, 2013. 8(5): p. 785-806.
152
28. Wang, C., A. Mbi, and M. Shepherd, A study on breath acetone in diabetic patients using
a cavity ringdown breath analyzer: exploring correlations of breath acetone with blood
glucose and glycohemoglobin A1C. IEEE Sensors Journal, 2009. 10(1): p. 54-63.
29. Ruzsányi, V. and M.P. Kalapos, Breath acetone as a potential marker in clinical practice.
Journal of breath research, 2017. 11(2): p. 024002.
30. Kneepkens, C.F., G. Lepage, and C.C. Roy, The potential of the hydrocarbon breath test
as a measure of lipid peroxidation. Free Radical Biology and Medicine, 1994. 17(2): p.
127-160.
31. Phillips, M., J. Greenberg, and R.N. Cataneo, Effect of age on the profile of alkanes in
normal human breath. Free radical research, 2000. 33(1): p. 57-63.
32. Phillips, M., et al., Effect of age on the breath methylated alkane contour, a display of
apparent new markers of oxidative stress. Journal of Laboratory and Clinical Medicine,
2000. 136(3): p. 243-249.
33. Risby, T.H. and S.S. Sehnert, Clinical application of breath biomarkers of oxidative stress
status. Free Radical Biology and Medicine, 1999. 27(11-12): p. 1182-1192.
34. de Lacy Costello, B., et al., A review of the volatiles from the healthy human body.
Journal of breath research, 2014. 8(1): p. 014001.
35. Drabińska, N., et al., A literature survey of all volatiles from healthy human breath and
bodily fluids: the human volatilome. Journal of breath research, 2021. 15(3): p. 034001.
36. Phillips, M., et al., Variation in volatile organic compounds in the breath of normal
humans. Journal of Chromatography B: Biomedical Sciences and Applications, 1999.
729(1-2): p. 75-88.
37. Alving, K., E. Weitzberg, and J. Lundberg, Increased amount of nitric oxide in exhaled air
of asthmatics. European Respiratory Journal, 1993. 6(9): p. 1368-1370.
38. Kharitonov, S., et al., Increased nitric oxide in exhaled air of asthmatic patients. The
Lancet, 1994. 343(8890): p. 133-135.
39. Persson, M.G., et al., Single-breath nitric oxide measurements in asthmatic patients and
smokers. The Lancet, 1994. 343(8890): p. 146-147.
40. Massaro, A.F., et al., Expired nitric oxide levels during treatment of acute asthma.
American journal of respiratory and critical care medicine, 1995. 152(2): p. 800-803.
41. Robbins, R.A., et al., Measurement of exhaled nitric oxide by three different techniques.
American journal of respiratory and critical care medicine, 1996. 153(5): p. 1631-1635.
153
42. Lee, Y., B.K. Oh, and M.E. Meyerhoff, Improved planar amperometric nitric oxide sensor
based on platinized platinum anode. 1. Experimental results and theory when applied for
monitoring NO release from diazeniumdiolate-doped polymeric films. Analytical
chemistry, 2004. 76(3): p. 536-544.
43. Privett, B.J., J.H. Shin, and M.H. Schoenfisch, Electrochemical nitric oxide sensors for
physiological measurements. Chemical Society Reviews, 2010. 39(6): p. 1925-1935.
44. Brown, M.D. and M.H. Schoenfisch, Electrochemical nitric oxide sensors: principles of
design and characterization. Chemical reviews, 2019. 119(22): p. 11551-11575.
45. Baharuddin, A.A., et al., Advances in chemiresistive sensors for acetone gas detection.
Materials science in semiconductor processing, 2019. 103: p. 104616.
46. Diskin, A.M., P. Španěl, and D. Smith, Time variation of ammonia, acetone, isoprene and
ethanol in breath: a quantitative SIFT-MS study over 30 days. Physiological
measurement, 2003. 24(1): p. 107.
47. Deng, C., et al., Determination of acetone in human breath by gas chromatography–
mass spectrometry and solid-phase microextraction with on-fiber derivatization. Journal
of Chromatography B, 2004. 810(2): p. 269-275.
48. Hu, J., et al., Synthesis and gas sensing properties of NiO/SnO2 hierarchical structures
toward ppb-level acetone detection. Materials Research Bulletin, 2018. 102: p. 294-303.
49. Righettoni, M., A. Tricoli, and S.E. Pratsinis, Si: WO3 sensors for highly selective detection
of acetone for easy diagnosis of diabetes by breath analysis. Analytical chemistry, 2010.
82(9): p. 3581-3587.
50. Righettoni, M., et al., Breath acetone monitoring by portable Si: WO3 gas sensors.
Analytica chimica acta, 2012. 738: p. 69-75.
51. Li, G., et al., Bimetal PdAu decorated SnO2 nanosheets based gas sensor with
temperature-dependent dual selectivity for detecting formaldehyde and acetone.
Sensors and Actuators B: Chemical, 2019. 283: p. 590-601.
52. Choi, S.-J., et al., Selective detection of acetone and hydrogen sulfide for the diagnosis of
diabetes and halitosis using SnO2 nanofibers functionalized with reduced graphene oxide
nanosheets. ACS applied materials & interfaces, 2014. 6(4): p. 2588-2597.
53. Choi, S.-J., et al., Selective diagnosis of diabetes using Pt-functionalized WO3 hemitube
networks as a sensing layer of acetone in exhaled breath. Analytical chemistry, 2013.
85(3): p. 1792-1796.
54. Amiri, V., et al., Nanostructured metal oxide-based acetone gas sensors: A review.
Sensors, 2020. 20(11): p. 3096.
154
55. Spichiger-Keller, U.E., Chemical sensors and biosensors for medical and biological
applications. 2008: John Wiley & Sons.
56. Kharitonov, S., K. Alving, and P. Barnes, Exhaled and nasal nitric oxide measurements:
recommendations. The European Respiratory Society task force. European Respiratory
Journal, 1997. 10(7): p. 1683-1693.
57. Maziak, W., et al., Exhaled nitric oxide in chronic obstructive pulmonary disease.
American journal of respiratory and critical care medicine, 1998. 157(3): p. 998-1002.
58. Pavord, I.D., et al., Inflammometry to assess airway diseases. The Lancet, 2008.
372(9643): p. 1017-1019.
59. Dotsch, J., et al., Airway nitric oxide in asthmatic children and patients with cystic
fibrosis. European Respiratory Journal, 1996. 9(12): p. 2537-2540.
60. Das, S. and M. Pal, Non-invasive monitoring of human health by exhaled breath analysis:
A comprehensive review. Journal of The Electrochemical Society, 2020. 167(3): p.
037562.
61. Issitt, T., et al., Volatile compounds in human breath: critical review and meta-analysis.
Journal of Breath Research, 2022.
62. Montuschi, P., et al., Diagnostic performance of an electronic nose, fractional exhaled
nitric oxide, and lung function testing in asthma. Chest, 2010. 137(4): p. 790-796.
63. Nelson, N., et al., Exhaled isoprene and acetone in newborn infants and in children with
diabetes mellitus. Pediatric research, 1998. 44(3): p. 363-367.
64. Mendis, S., P.A. Sobotka, and D.E. Euler, Pentane and isoprene in expired air from
humans: gas-chromatographic analysis of single breath. Clinical Chemistry, 1994. 40(8):
p. 1485-1488.
65. Sparkman, O.D., Z. Penton, and F.G. Kitson, Gas chromatography and mass
spectrometry: a practical guide. 2011: Academic press.
66. Lisec, J., et al., Gas chromatography mass spectrometry–based metabolite profiling in
plants. Nature protocols, 2006. 1(1): p. 387-396.
67. Pasikanti, K.K., P. Ho, and E. Chan, Gas chromatography/mass spectrometry in metabolic
profiling of biological fluids. Journal of Chromatography B, 2008. 871(2): p. 202-211.
68. Jansson, B. and B. Larsson, Analysis of organic compounds in human breath by gas
chromatography-mass spectrometry. The Journal of laboratory and clinical medicine,
1969. 74(6): p. 961-966.
155
69. Gohlke, R.S., Time-of-flight mass spectrometry and gas-liquid partition chromatography.
Analytical Chemistry, 1959. 31(4): p. 535-541.
70. Horning, E. and M. Horning, Human metabolic profiles obtained by GC and GC/MS.
Journal of Chromatographic Science, 1971. 9(3): p. 129-140.
71. Horning, E.C. and M.-G. Horning, Metabolic profiles: gas-phase methods for analysis of
metabolites. Clinical chemistry, 1971. 17(8): p. 802-809.
72. Zlatkis, A., et al., Profile of volatile metabolites in urine by gas chromatography-mass
spectrometry. Analytical chemistry, 1973. 45(4): p. 763-767.
73. Zlatkis, A. and H.M. Liebich, Profile of volatile metabolites in human urine. Clinical
Chemistry, 1971. 17(7): p. 592-594.
74. Pauling, L., et al., Quantitative analysis of urine vapor and breath by gas-liquid partition
chromatography. Proceedings of the National Academy of Sciences, 1971. 68(10): p.
2374-2376.
75. Hiroshi, K., et al., Evaluation of volatile sulfur compounds in the expired alveolar gas in
patients with liver cirrhosis. Clinica Chimica Acta, 1978. 85(3): p. 279-284.
76. Gordon, S., et al., Volatile organic compounds in exhaled air from patients with lung
cancer. Clinical chemistry, 1985. 31(8): p. 1278-1282.
77. O'neill, H., et al., A computerized classification technique for screening for the presence
of breath biomarkers in lung cancer. Clinical chemistry, 1988. 34(8): p. 1613-1618.
78. Phillips, M., et al., Volatile organic compounds in breath as markers of lung cancer: a
cross-sectional study. The Lancet, 1999. 353(9168): p. 1930-1933.
79. Phillips, M., M. Sabas, and J. Greenberg, Increased pentane and carbon disulfide in the
breath of patients with schizophrenia. Journal of clinical pathology, 1993. 46(9): p. 861-
864.
80. Phillips, M., et al., Volatile organic compounds in the breath of patients with
schizophrenia. Journal of clinical pathology, 1995. 48(5): p. 466-469.
81. Pleil, J.D. and A.B. Lindstrom, Measurement of volatile organic compounds in exhaled
breath as collected in evacuated electropolished canisters. Journal of Chromatography B:
Biomedical Sciences and Applications, 1995. 665(2): p. 271-279.
82. Phillips, M., Method for the collection and assay of volatile organic compounds in breath.
Analytical biochemistry, 1997. 247(2): p. 272-278.
156
83. Grote, C. and J. Pawliszyn, Solid-phase microextraction for the analysis of human breath.
Analytical chemistry, 1997. 69(4): p. 587-596.
84. Baltussen, E., C. Cramers, and P. Sandra, Sorptive sample preparation–a review.
Analytical and bioanalytical chemistry, 2002. 373(1): p. 3-22.
85. Woolfenden, E., Optimising analytical performance and extending the application range
of thermal desorption for indoor air monitoring. Indoor and Built Environment, 2001.
10(3-4): p. 222-231.
86. Henderson, K. and I. Matthews, Biological monitoring of midwives' exposure to N2O
using the Bio-VOC breath sampler. Journal of Exposure Science & Environmental
Epidemiology, 2002. 12(5): p. 309-312.
87. Poli, D., et al., Determination of aldehydes in exhaled breath of patients with lung cancer
by means of on-fiber-derivatisation SPME–GC/MS. Journal of Chromatography B, 2010.
878(27): p. 2643-2651.
88. Gashimova, E., et al., Investigation of different approaches for exhaled breath and tumor
tissue analyses to identify lung cancer biomarkers. Heliyon, 2020. 6(6): p. e04224.
89. Schubert, J.K., et al., CO2-controlled sampling of alveolar gas in mechanically ventilated
patients. Journal of Applied Physiology, 2001. 90(2): p. 486-492.
90. Miekisch, W., et al., Impact of sampling procedures on the results of breath analysis.
Journal of breath research, 2008. 2(2): p. 026007.
91. Horváth, I., et al., A European Respiratory Society technical standard: exhaled
biomarkers in lung disease. European Respiratory Journal, 2017. 49(4).
92. Aghdassi, E., et al., Antioxidant vitamin supplementation in Crohn’s disease decreases
oxidative stress: a randomized controlled trial. The American journal of
gastroenterology, 2003. 98(2): p. 348-353.
93. Van den Velde, S., et al., GC–MS analysis of breath odor compounds in liver patients.
Journal of Chromatography B, 2008. 875(2): p. 344-348.
94. Xue, R., et al., Investigation of volatile biomarkers in liver cancer blood using solid-phase
microextraction and gas chromatography/mass spectrometry. Rapid Communications in
Mass Spectrometry: An International Journal Devoted to the Rapid Dissemination of Up-
to-the-Minute Research in Mass Spectrometry, 2008. 22(8): p. 1181-1186.
95. Altomare, D., et al., Exhaled volatile organic compounds identify patients with colorectal
cancer. Journal of British Surgery, 2013. 100(1): p. 144-150.
157
96. Altomare, D.F., et al., Effects of curative colorectal cancer surgery on exhaled volatile
organic compounds and potential implications in clinical follow-up. Annals of Surgery,
2015. 262(5): p. 862-867.
97. Altomare, D., et al., Chemical signature of colorectal cancer: Case–control study for
profiling the breath print. BJS open, 2020. 4(6): p. 1189-1199.
98. Wang, C., et al., Volatile organic metabolites identify patients with breast cancer,
cyclomastopathy and mammary gland fibroma. Scientific reports, 2014. 4(1): p. 1-6.
99. Li, J., et al., Investigation of potential breath biomarkers for the early diagnosis of breast
cancer using gas chromatography–mass spectrometry. Clinica chimica acta, 2014. 436:
p. 59-67.
100. Phillips, M., et al., Rapid point-of-care breath test predicts breast cancer and abnormal
mammograms in symptomatic women. MedRxiv, 2020.
101. Xu, Z., et al., A nanomaterial-based breath test for distinguishing gastric cancer from
benign gastric conditions. British journal of cancer, 2013. 108(4): p. 941-950.
102. Amal, H., et al., Detection of precancerous gastric lesions and gastric cancer through
exhaled breath. Gut, 2016. 65(3): p. 400-407.
103. Chen, Y., et al., Breath analysis based on surface-enhanced Raman scattering sensors
distinguishes early and advanced gastric cancer patients from healthy persons. ACS
nano, 2016. 10(9): p. 8169-8179.
104. Tong, H., et al., Volatile organic metabolites identify patients with gastric carcinoma,
gastric ulcer, or gastritis and control patients. Cancer Cell International, 2017. 17(1): p.
1-9.
105. Mochalski, P., et al., Ex vivo emission of volatile organic compounds from gastric cancer
and non-cancerous tissue. Journal of breath research, 2018. 12(4): p. 046005.
106. Gruber, M., et al., Analysis of exhaled breath for diagnosing head and neck squamous
cell carcinoma: a feasibility study. British journal of cancer, 2014. 111(4): p. 790-798.
107. García, R.A., et al., Volatile organic compounds analysis in breath air in healthy
volunteers and patients suffering epidermoid laryngeal carcinomas. Chromatographia,
2014. 77(5): p. 501-509.
108. Guo, L., et al., Exhaled breath volatile biomarker analysis for thyroid cancer.
Translational Research, 2015. 166(2): p. 188-195.
109. Amal, H., et al., Assessment of ovarian cancer conditions from exhaled breath.
International journal of cancer, 2015. 136(6): p. E614-E622.
158
110. Kolk, A., et al., Breath analysis as a potential diagnostic tool for tuberculosis. The
International Journal of Tuberculosis and Lung Disease, 2012. 16(6): p. 777-782.
111. Amann, A., et al., Applications of breath gas analysis in medicine. International Journal
of Mass Spectrometry, 2004. 239(2-3): p. 227-233.
112. Manolis, A., The diagnostic potential of breath analysis. Clinical chemistry, 1983. 29(1):
p. 5-15.
113. Nardi-Agmon, I. and N. Peled, Exhaled breath analysis for the early detection of lung
cancer: recent developments and future prospects. Lung Cancer: Targets and Therapy,
2017. 8: p. 31.
114. Nakhleh, M.K., et al., Diagnosis and classification of 17 diseases from 1404 subjects via
pattern analysis of exhaled molecules. ACS nano, 2017. 11(1): p. 112-125.
115. Paredi, P., et al., Exhaled ethane is elevated in cystic fibrosis and correlates with carbon
monoxide levels and airway obstruction. American journal of respiratory and critical care
medicine, 2000. 161(4): p. 1247-1251.
116. Peled, N., et al., Non-invasive breath analysis of pulmonary nodules. Journal of Thoracic
Oncology, 2012. 7(10): p. 1528-1533.
117. Capuano, R., et al., The lung cancer breath signature: a comparative analysis of exhaled
breath and air sampled from inside the lungs. Scientific Reports, 2015. 5(1): p. 1-10.
118. Lamote, K., et al., Breath analysis by gas chromatography-mass spectrometry and
electronic nose to screen for pleural mesothelioma: a cross-sectional case-control study.
Oncotarget, 2017. 8(53): p. 91593.
119. Wang, C., et al., Exhaled volatile organic compounds as lung cancer biomarkers during
one-lung ventilation. Scientific reports, 2014. 4(1): p. 1-8.
120. Filipiak, W., et al., Comparative analyses of volatile organic compounds (VOCs) from
patients, tumors and transformed cell lines for the validation of lung cancer-derived
breath markers. Journal of breath research, 2014. 8(2): p. 027111.
121. Corradi, M., et al., Exhaled breath analysis in suspected cases of non-small-cell lung
cancer: a cross-sectional study. Journal of breath research, 2015. 9(2): p. 027101.
122. Schallschmidt, K., et al., Comparison of volatile organic compounds from lung cancer
patients and healthy controls—Challenges and limitations of an observational study.
Journal of breath research, 2016. 10(4): p. 046007.
123. Oguma, T., et al., Clinical contributions of exhaled volatile organic compounds in the
diagnosis of lung cancer. PloS one, 2017. 12(4): p. e0174802.
159
124. Callol-Sanchez, L., et al., Observation of nonanoic acid and aldehydes in exhaled breath
of patients with lung cancer. Journal of breath research, 2017. 11(2): p. 026004.
125. Sakumura, Y., et al., Diagnosis by volatile organic compounds in exhaled breath from
lung cancer patients using support vector machine algorithm. Sensors, 2017. 17(2): p.
287.
126. Saidi, T., et al., Non-invasive prediction of lung cancer histological types through exhaled
breath analysis by UV-irradiated electronic nose and GC/QTOF/MS. Sensors and
Actuators B: Chemical, 2020. 311: p. 127932.
127. Koureas, M., et al., Target analysis of volatile organic compounds in exhaled breath for
lung cancer discrimination from other pulmonary diseases and healthy persons.
Metabolites, 2020. 10(8): p. 317.
128. Di Gilio, A., et al., Breath analysis for early detection of malignant pleural mesothelioma:
volatile organic compounds (VOCs) determination and possible biochemical pathways.
Cancers, 2020. 12(5): p. 1262.
129. Sponring, A., et al., Release of volatile organic compounds from the lung cancer cell line
NCI-H2087 in vitro. Anticancer research, 2009. 29(1): p. 419-426.
130. Filipiak, W., et al., Release of volatile organic compounds (VOCs) from the lung cancer
cell line CALU-1 in vitro. Cancer cell international, 2008. 8(1): p. 1-11.
131. Filipiak, W., et al., TD-GC-MS analysis of volatile metabolites of human lung cancer and
normal cells in vitro. Cancer epidemiology, biomarkers & prevention, 2010. 19(1): p.
182-195.
132. Deng, C., X. Zhang, and N. Li, Investigation of volatile biomarkers in lung cancer blood
using solid-phase microextraction and capillary gas chromatography–mass
spectrometry. Journal of Chromatography B, 2004. 808(2): p. 269-277.
133. Ma, H., et al., Analysis of human breath samples of lung cancer patients and healthy
controls with solid-phase microextraction (SPME) and flow-modulated comprehensive
two-dimensional gas chromatography (GC× GC). Analytical Methods, 2014. 6(17): p.
6841-6849.
134. Barash, O., et al., Differentiation between genetic mutations of breast cancer by breath
volatolomics. Oncotarget, 2015. 6(42): p. 44864.
135. Kalluri, U., M. Naiker, and M. Myers, Cell culture metabolomics in the diagnosis of lung
cancer—the influence of cell culture conditions. Journal of breath research, 2014. 8(2): p.
027109.
160
136. Filipiak, W., et al., A compendium of volatile organic compounds (VOCs) released by
human cell lines. Current medicinal chemistry, 2016. 23(20): p. 2112-2131.
137. Sponring, A., et al., Analysis of volatile organic compounds (VOCs) in the headspace of
NCI-H1666 lung cancer cells. Cancer Biomarkers, 2010. 7(3): p. 153-161.
138. Barash, O., et al., Classification of lung cancer histology by gold nanoparticle sensors.
Nanomedicine: Nanotechnology, Biology and Medicine, 2012. 8(5): p. 580-589.
139. Rutter, A.V., et al., Quantification by SIFT-MS of acetaldehyde released by lung cells in a
3D model. Analyst, 2013. 138(1): p. 91-95.
140. Peled, N., et al., Volatile fingerprints of cancer specific genetic mutations. Nanomedicine:
Nanotechnology, Biology and Medicine, 2013. 9(6): p. 758-766.
141. Davies, M., et al., Unique volatolomic signatures of TP53 and KRAS in lung cells. British
journal of cancer, 2014. 111(6): p. 1213-1221.
142. Schallschmidt, K., et al., Investigation of cell culture volatilomes using solid phase micro
extraction: Options and pitfalls exemplified with adenocarcinoma cell lines. Journal of
chromatography B, 2015. 1006: p. 158-166.
143. Thriumani, R., et al., A Study on VOCs Released by Lung Cancer Cell Line Using GCMS-
SPME. Procedia Chemistry, 2016. 20: p. 1-7.
144. Jia, Z., et al., Detection of lung cancer: concomitant volatile organic compounds and
metabolomic profiling of six cancer cell lines of different histological origins. ACS omega,
2018. 3(5): p. 5131-5140.
145. Thriumani, R., et al., A study on volatile organic compounds emitted by in-vitro lung
cancer cultured cells using gas sensor array and SPME-GCMS. BMC cancer, 2018. 18(1):
p. 1-17.
146. Janssens, E., J.P. van Meerbeeck, and K. Lamote, Volatile organic compounds in human
matrices as lung cancer biomarkers: a systematic review. Critical Reviews in
Oncology/Hematology, 2020. 153: p. 103037.
147. Janssens, E., et al., Headspace Volatile Organic Compound Profiling of Pleural
Mesothelioma and Lung Cancer Cell Lines as Translational Bridge for Breath Research.
Frontiers in Oncology, 2022. 12.
148. Huang, J., et al., Selected ion flow tube mass spectrometry analysis of volatile
metabolites in urine headspace for the profiling of gastro-esophageal cancer. Analytical
Chemistry, 2013. 85(6): p. 3409-3416.
161
149. He, J., et al., Fingerprinting breast cancer vs. normal mammary cells by mass
spectrometric analysis of volatiles. Scientific reports, 2014. 4(1): p. 1-6.
150. Lavra, L., et al., Investigation of VOCs associated with different characteristics of breast
cancer cells. Scientific reports, 2015. 5(1): p. 1-12.
151. Silva, C.L., et al., Volatile metabolomic signature of human breast cancer cell lines.
Scientific reports, 2017. 7(1): p. 1-8.
152. Tang, H., et al., Determination of volatile organic compounds exhaled by cell lines
derived from hematological malignancies. Bioscience reports, 2017. 37(3).
153. Forleo, A., et al. Evaluation of the volatile organic compounds released from peripheral
blood mononuclear cells and THP1 cells under normal and proinflammatory conditions.
in AISEM Annual Conference on Sensors and Microsystems. 2017. Springer.
154. Amal, H., et al., The scent fingerprint of hepatocarcinoma: in-vitro metastasis prediction
with volatile organic compounds (VOCs). International journal of nanomedicine, 2012. 7:
p. 4135.
155. Mochalski, P., et al., Release and uptake of volatile organic compounds by human
hepatocellular carcinoma cells (HepG2) in vitro. Cancer cell international, 2013. 13(1): p.
1-9.
156. Lima, A.R., et al., Discrimination between the human prostate normal and cancer cell
exometabolome by GC-MS. Scientific Reports, 2018. 8(1): p. 1-12.
157. Phillips, M., et al., Prediction of lung cancer using volatile biomarkers in breath. Cancer
biomarkers, 2007. 3(2): p. 95-109.
158. Phillips, M., et al., Detection of lung cancer with volatile markers in the breath. Chest,
2003. 123(6): p. 2115-2123.
159. Phillips, M., et al., Blinded validation of breath biomarkers of lung cancer, a potential
ancillary to chest CT screening. PLoS One, 2015. 10(12): p. e0142484.
160. McCunney, R.J. and J. Li, Radiation risks in lung cancer screening programs. Chest, 2014.
145(3): p. 618-624.
161. Maziak, D.E., et al., Positron emission tomography in staging early lung cancer: a
randomized trial. Annals of internal medicine, 2009. 151(4): p. 221-228.
162. Vansteenkiste, J., et al., Screening and early—detection of lung cancer. Annals of
Oncology, 2012. 23: p. x320-x327.
162
163. van Klaveren, R.J., et al., Management of lung nodules detected by volume CT scanning.
New England Journal of Medicine, 2009. 361(23): p. 2221-2229.
164. Fang, M., et al., Thermal degradation of small molecules: a global metabolomic
investigation. Analytical chemistry, 2015. 87(21): p. 10935-10941.
165. Hauschild, A.-C., J.I. Baumbach, and J. Baumbach, Integrated statistical learning of
metabolic ion mobility spectrometry profiles for pulmonary disease identification. Genet.
Mol. Res, 2012. 11(3): p. 2733-2744.
166. Westhoff, M., et al., Differentiation of chronic obstructive pulmonary disease (COPD)
including lung cancer from healthy control group by breath analysis using ion mobility
spectrometry. International Journal for Ion Mobility Spectrometry, 2010. 13(3): p. 131-
139.
167. Basanta, M., et al., Non-invasive metabolomic analysis of breath using differential
mobility spectrometry in patients with chronic obstructive pulmonary disease and
healthy smokers. Analyst, 2010. 135(2): p. 315-320.
168. Borsdorf, H. and G.A. Eiceman, Ion mobility spectrometry: principles and applications.
Applied Spectroscopy Reviews, 2006. 41(4): p. 323-375.
169. Fernández-Maestre, R., Ion mobility spectrometry: history, characteristics and
applications. Revista UDCA Actualidad & Divulgación Científica, 2012. 15(2): p. 467-479.
170. Merenbloom, S.I., et al., High-resolution ion cyclotron mobility spectrometry. Analytical
chemistry, 2009. 81(4): p. 1482-1487.
171. Eiceman, G.A. and Z. Karpas, Ion mobility spectrometry. 2005: CRC press.
172. Cumeras, R., et al., Review on ion mobility spectrometry. Part 1: current instrumentation.
Analyst, 2015. 140(5): p. 1376-1390.
173. Covington, J., et al., The application of FAIMS gas analysis in medical diagnostics.
Analyst, 2015. 140(20): p. 6775-6781.
174. Kalkan, Y., et al., Cluster ions in gas-based detectors. Journal of Instrumentation, 2015.
10(07): p. P07004.
175. Ruzsanyi, V., et al., Detection of human metabolites using multi-capillary columns
coupled to ion mobility spectrometers. Journal of Chromatography A, 2005. 1084(1-2): p.
145-151.
176. Fink, T., J.I. Baumbach, and S. Kreuer, Ion mobility spectrometry in breath research.
Journal of breath research, 2014. 8(2): p. 027104.
163
177. Westhoff, M., et al., Ion mobility spectrometry: a new method for the detection of lung
cancer and airway infection in exhaled air? First results of a pilot study. Chest, 2005.
128(4): p. 155S.
178. Baumbach, J. and M. Westhoff, Ion mobility spectrometry to detect lung cancer and
airway infections. Spectroscopy Europe, 2006. 18(6): p. 22-27.
179. Darwiche, K., et al., Bronchoscopically obtained volatile biomarkers in lung cancer. Lung,
2011. 189(6): p. 445-452.
180. Baumbach, J., et al., Significant different volatile biomarker during bronchoscopic ion
mobility spectrometry investigation of patients suffering lung carcinoma. International
Journal for Ion Mobility Spectrometry, 2011. 14(4): p. 159-166.
181. Bessa, V., et al., Detection of volatile organic compounds (VOCs) in exhaled breath of
patients with chronic obstructive pulmonary disease (COPD) by ion mobility
spectrometry. International Journal for Ion Mobility Spectrometry, 2011. 14(1): p. 7-13.
182. Westhoff, M., et al., Statistical and bioinformatical methods to differentiate chronic
obstructive pulmonary disease (COPD) including lung cancer from healthy control by
breath analysis using ion mobility spectrometry. International Journal for Ion Mobility
Spectrometry, 2011. 14(4): p. 139-149.
183. Rabis, T., et al., Detection of infectious agents in the airways by ion mobility
spectrometry of exhaled breath. International Journal for Ion Mobility Spectrometry,
2011. 14(4): p. 187-195.
184. Maddula, S., et al., Correlation analysis on data sets to detect infectious agents in the
airways by ion mobility spectrometry of exhaled breath. International Journal for Ion
Mobility Spectrometry, 2011. 14(4): p. 197-206.
185. Pagonas, N., et al., Volatile organic compounds in uremia. 2012.
186. Furtwängler, R., et al., Signals of neutropenia in human breath? International journal for
ion mobility spectrometry, 2014. 17(1): p. 19-23.
187. Bunkowski, A., et al., MCC/IMS signals in human breath related to sarcoidosis—results of
a feasibility study using an automated peak finding procedure. Journal of Breath
Research, 2009. 3(4): p. 046001.
188. Koczulla, R., et al., MCC/IMS as potential noninvasive technique in the diagnosis of
patients with COPD with and without alpha 1-antitrypsin deficiency. International
Journal for Ion Mobility Spectrometry, 2011. 14(4): p. 177-185.
189. Arasaradnam, R.P., et al., Detection of colorectal cancer (CRC) by urinary volatile organic
compound analysis. PloS one, 2014. 9(9): p. e108750.
164
190. Niemi, R.J., et al., FAIMS analysis of urine gaseous headspace is capable of
differentiating ovarian cancer. Gynecologic Oncology, 2018. 151(3): p. 519-524.
191. Arasaradnam, R.P., et al., Noninvasive diagnosis of pancreatic cancer through detection
of volatile organic compounds in urine. Gastroenterology, 2018. 154(3): p. 485-487. e1.
192. Westhoff, M., et al., Ion mobility spectrometry for the detection of volatile organic
compounds in exhaled breath of patients with lung cancer: results of a pilot study.
Thorax, 2009. 64(9): p. 744-748.
193. Miekisch, W. and J.K. Schubert, From highly sophisticated analytical techniques to life-
saving diagnostics: Technical developments in breath analysis. TrAC Trends in Analytical
Chemistry, 2006. 25(7): p. 665-673.
194. Costanzo, M.T., et al., Portable FAIMS: Applications and future perspectives.
International journal of mass spectrometry, 2017. 422: p. 188-196.
195. Vautz, W., et al., Breath analysis—performance and potential of ion mobility
spectrometry. Journal of breath research, 2009. 3(3): p. 036004.
196. Neri, G., et al., Real-time monitoring of breath ammonia during haemodialysis: use of ion
mobility spectrometry (IMS) and cavity ring-down spectroscopy (CRDS) techniques.
Nephrology Dialysis Transplantation, 2012. 27(7): p. 2945-2952.
197. Vautz, W., et al., On the potential of ion mobility spectrometry coupled to GC pre-
separation–A tutorial. Analytica chimica acta, 2018. 1024: p. 52-64.
198. Vautz, W., S. Sielemann, and J.I. Baumbach, Determination of terpenes in humid ambient
air using ultraviolet ion mobility spectrometry. Analytica Chimica Acta, 2004. 513(2): p.
393-399.
199. Vautz, W., et al., Sensitive ion mobility spectrometry of humid ambient air using 10.6 eV
UV-IMS. Int. J. Ion Mobility Spectrom, 2004. 7(3).
200. Dummer, J., et al., Analysis of biogenic volatile organic compounds in human health and
disease. TrAC Trends in Analytical Chemistry, 2011. 30(7): p. 960-967.
201. Baumbach, J.I., Ion mobility spectrometry coupled with multi-capillary columns for
metabolic profiling of human breath. Journal of breath research, 2009. 3(3): p. 034001.
202. Kirk, A.T., et al., Ultra-high-resolution ion mobility spectrometry—current
instrumentation, limitations, and future developments. Analytical and bioanalytical
chemistry, 2019. 411(24): p. 6229-6246.
203. Thaler, E.R. and C.W. Hanson, Medical applications of electronic nose technology. Expert
review of medical devices, 2005. 2(5): p. 559-566.
165
204. Turner, A.P. and N. Magan, Electronic noses and disease diagnostics. Nature Reviews
Microbiology, 2004. 2(2): p. 161-166.
205. Jia, Z., et al., Critical review of volatile organic compound analysis in breath and in vitro
cell culture for detection of lung cancer. Metabolites, 2019. 9(3): p. 52.
206. Amal, H. and H. Haick, Point of care breath analysis systems, in Advanced Nanomaterials
for Inexpensive Gas Microsensors. 2020, Elsevier. p. 315-334.
207. Konvalina, G. and H. Haick, Sensors for breath testing: from nanomaterials to
comprehensive disease detection. Accounts of chemical research, 2014. 47(1): p. 66-76.
208. Guntner, A.T., et al., Breath sensors for health monitoring. ACS sensors, 2019. 4(2): p.
268-280.
209. Cellini, A., et al., Potential applications and limitations of electronic nose devices for
plant disease diagnosis. Sensors, 2017. 17(11): p. 2596.
210. Persaud, K. and G. Dodd, Analysis of discrimination mechanisms in the mammalian
olfactory system using a model nose. Nature, 1982. 299(5881): p. 352-355.
211. Berna, A., Metal oxide sensors for electronic noses and their application to food analysis.
Sensors, 2010. 10(4): p. 3882-3910.
212. Kononov, A., et al., Online breath analysis using metal oxide semiconductor sensors
(electronic nose) for diagnosis of lung cancer. Journal of breath research, 2019. 14(1): p.
016004.
213. Blatt, R., et al. Lung cancer identification by an electronic nose based on an array of MOS
sensors. in 2007 International joint conference on neural networks. 2007. IEEE.
214. van Hooren, M.R., et al., Differentiating head and neck carcinoma from lung carcinoma
with an electronic nose: a proof of concept study. European Archives of Oto-Rhino-
Laryngology, 2016. 273(11): p. 3897-3903.
215. Kort, S., et al., Multi-centre prospective study on diagnosing subtypes of lung cancer by
exhaled-breath analysis. Lung Cancer, 2018. 125: p. 223-229.
216. Van de Goor, R., et al., Training and validating a portable electronic nose for lung cancer
screening. Journal of Thoracic Oncology, 2018. 13(5): p. 676-681.
217. De Vries, R., et al., Integration of electronic nose technology with spirometry: validation
of a new approach for exhaled breath analysis. Journal of breath research, 2015. 9(4): p.
046001.
166
218. Marzorati, D., et al. A Metal Oxide Gas Sensors Array for Lung Cancer Diagnosis Through
Exhaled Breath Analysis. in 2019 41st Annual International Conference of the IEEE
Engineering in Medicine and Biology Society (EMBC). 2019. IEEE.
219. Leunis, N., et al., Application of an electronic nose in the diagnosis of head and neck
cancer. The Laryngoscope, 2014. 124(6): p. 1377-1381.
220. Van De Goor, R., et al., Feasibility of electronic nose technology for discriminating
between head and neck, bladder, and colon carcinomas. European Archives of Oto-
Rhino-Laryngology, 2017. 274(2): p. 1053-1060.
221. van de Goor, R.M., et al., Detecting recurrent head and neck cancer using electronic nose
technology: a feasibility study. Head & Neck, 2019. 41(9): p. 2983-2990.
222. Schuermans, V.N., et al., Pilot study: detection of gastric cancer from exhaled air
analyzed with an electronic nose in Chinese patients. Surgical innovation, 2018. 25(5): p.
429-434.
223. Roine, A., et al., Detection of prostate cancer by an electronic nose: a proof of principle
study. The Journal of urology, 2014. 192(1): p. 230-235.
224. Waltman, C.G., T.A. Marcelissen, and J.G. van Roermund, Exhaled-breath testing for
prostate cancer based on volatile organic compound profiling using an electronic nose
device (Aeonose™): a preliminary report. European urology focus, 2020. 6(6): p. 1220-
1225.
225. Horvath, G., J. Chilo, and T. Lindblad, Different volatile signals emitted by human ovarian
carcinoma and healthy tissue. Future Oncology, 2010. 6(6): p. 1043-1049.
226. Weber, C.M., et al., Evaluation of a gas sensor array and pattern recognition for the
identification of bladder cancer from urine headspace. Analyst, 2011. 136(2): p. 359-364.
227. Horstmann, M., et al., PD25-03 an electronic nose system detects bladder cancer in urine
specimen: First results of a pilot study. The Journal of Urology, 2015. 193(4S): p. e560-
e561.
228. Guo, D., et al., A novel breath analysis system based on electronic olfaction. IEEE
transactions on biomedical engineering, 2010. 57(11): p. 2753-2763.
229. Ping, W., et al., A novel method for diabetes diagnosis based on electronic nose.
Biosensors and bioelectronics, 1997. 12(9-10): p. 1031-1036.
230. Voss, A., et al., Smelling renal dysfunction via electronic nose. Annals of biomedical
engineering, 2005. 33(5): p. 656-660.
167
231. Baldini, C., et al., Electronic nose as a novel method for diagnosing cancer: a systematic
review. Biosensors, 2020. 10(8): p. 84.
232. Schaller, E., J.O. Bosset, and F. Escher, ‘Electronic noses’ and their application to food.
LWT-Food Science and Technology, 1998. 31(4): p. 305-316.
233. Lewis, N.S., Comparisons between mammalian and artificial olfaction based on arrays of
carbon black− polymer composite vapor detectors. Accounts of chemical research, 2004.
37(9): p. 663-672.
234. Wilson, A.D. and M. Baietto, Applications and advances in electronic-nose technologies.
Sensors, 2009. 9(7): p. 5099-5148.
235. Machado, R.F., et al., Detection of lung cancer by sensor array analyses of exhaled
breath. American journal of respiratory and critical care medicine, 2005. 171(11): p.
1286-1291.
236. Tirzīte, M., et al., Detection of lung cancer in exhaled breath with an electronic nose
using support vector machine analysis. Journal of breath research, 2017. 11(3): p.
036009.
237. Tirzïte, M., et al., Detection of lung cancer with electronic nose and logistic regression
analysis. Journal of breath research, 2018. 13(1): p. 016006.
238. Dragonieri, S., et al., An electronic nose in the discrimination of patients with non-small
cell lung cancer and COPD. Lung cancer, 2009. 64(2): p. 166-170.
239. de Meij, T.G., et al., Electronic nose can discriminate colorectal carcinoma and advanced
adenomas by fecal volatile biomarker analysis: proof of principle study. International
journal of cancer, 2014. 134(5): p. 1132-1138.
240. Chapman, E.A., et al., A breath test for malignant mesothelioma using an electronic
nose. European Respiratory Journal, 2012. 40(2): p. 448-454.
241. Dragonieri, S., et al., An electronic nose distinguishes exhaled breath of patients with
Malignant Pleural Mesothelioma from controls. Lung cancer, 2012. 75(3): p. 326-331.
242. Bikov, A., et al., Expiratory flow rate, breath hold and anatomic dead space influence
electronic nose ability to detect lung cancer. BMC pulmonary medicine, 2014. 14(1): p. 1-
9.
243. Hubers, A.J., et al., Combined sputum hypermethylation and eNose analysis for lung
cancer diagnosis. Journal of Clinical Pathology, 2014. 67(8): p. 707-711.
168
244. McWilliams, A., et al., Sex and smoking status effects on the early detection of early lung
cancer in high-risk smokers using an electronic nose. IEEE Transactions on Biomedical
Engineering, 2015. 62(8): p. 2044-2054.
245. Yu, J.-B., et al., Analysis of diabetic patient's breath with conducting polymer sensor
array. Sensors and Actuators B: Chemical, 2005. 108(1-2): p. 305-308.
246. Fend, R., et al., Monitoring haemodialysis using electronic nose and chemometrics.
Biosensors and Bioelectronics, 2004. 19(12): p. 1581-1590.
247. Fens, N., et al., Exhaled breath profiling enables discrimination of chronic obstructive
pulmonary disease and asthma. American journal of respiratory and critical care
medicine, 2009. 180(11): p. 1076-1082.
248. Dragonieri, S., et al., An electronic nose in the discrimination of patients with asthma and
controls. Journal of allergy and clinical immunology, 2007. 120(4): p. 856-862.
249. Di Natale, C., et al., Lung cancer identification by the analysis of breath by means of an
array of non-selective gas sensors. Biosensors and Bioelectronics, 2003. 18(10): p. 1209-
1218.
250. D’Amico, A., et al., An investigation on electronic nose diagnosis of lung cancer. Lung
cancer, 2010. 68(2): p. 170-176.
251. Santonico, M., et al., In situ detection of lung cancer volatile fingerprints using
bronchoscopic air-sampling. Lung cancer, 2012. 77(1): p. 46-50.
252. Gasparri, R., et al., Volatile signature for the early diagnosis of lung cancer. Journal of
breath research, 2016. 10(1): p. 016007.
253. Di Natale, C., R. Paolesse, and A. D’Amico, Metalloporphyrins based artificial olfactory
receptors. Sensors and Actuators B: Chemical, 2007. 121(1): p. 238-246.
254. Bernabei, M., et al., A preliminary study on the possibility to diagnose urinary tract
cancers by an electronic nose. Sensors and Actuators B: Chemical, 2008. 131(1): p. 1-4.
255. D’Amico, A., et al., A novel approach for prostate cancer diagnosis using a gas sensor
array. Procedia Engineering, 2012. 47: p. 1113-1116.
256. Asimakopoulos, A., et al., Prostate cancer diagnosis through electronic nose in the urine
headspace setting: a pilot study. Prostate cancer and prostatic diseases, 2014. 17(2): p.
206-211.
257. Santonico, M., et al., Chemical sensors for prostate cancer detection oriented to non-
invasive approach. Procedia Engineering, 2014. 87: p. 320-323.
169
258. D'amico, A., et al., Identification of melanoma with a gas sensor array. Skin Research
and Technology, 2008. 14(2): p. 226-236.
259. Lin, Y.-J., et al., Application of the electronic nose for uremia diagnosis. Sensors and
Actuators B: Chemical, 2001. 76(1-3): p. 177-180.
260. Di Natale, C., et al., Electronic nose analysis of urine samples containing blood.
Physiological Measurement, 1999. 20(4): p. 377.
261. Haick, H., Chemical sensors based on molecularly modified metallic nanoparticles.
Journal of Physics D: Applied Physics, 2007. 40(23): p. 7173.
262. Tisch, U. and H. Haick, Nanomaterials for cross-reactive sensor arrays. MRS bulletin,
2010. 35(10): p. 797-803.
263. Peng, G., et al., Diagnosing lung cancer in exhaled breath using gold nanoparticles.
Nature Nanotechnology, 2009. 4(10): p. 669-673.
264. Hakim, M., et al., Diagnosis of head-and-neck cancer from exhaled breath. British journal
of cancer, 2011. 104(10): p. 1649-1655.
265. Peng, G., et al., Detection of lung, breast, colorectal, and prostate cancers from exhaled
breath using a single array of nanosensors. British journal of cancer, 2010. 103(4): p.
542-551.
266. Barash, O., et al., Sniffing the Unique “Odor Print” of Non-Small-Cell Lung Cancer with
Gold Nanoparticles. Small, 2009. 5(22): p. 2618-2624.
267. Shuster, G., et al., Classification of breast cancer precursors through exhaled breath.
Breast cancer research and treatment, 2011. 126(3): p. 791-796.
268. Wang, B., et al., Artificial sensing intelligence with silicon nanowires for ultraselective
detection in the gas phase. Nano letters, 2014. 14(2): p. 933-938.
269. Shehada, N., et al., Ultrasensitive silicon nanowire for real-world gas sensing:
noninvasive diagnosis of cancer from breath volatolome. Nano letters, 2015. 15(2): p.
1288-1295.
270. Zilberman, Y., et al., Carbon nanotube/hexa-peri-hexabenzocoronene bilayers for
discrimination between nonpolar volatile organic compounds of cancer and humid
atmospheres. Advanced Materials, 2010. 22(38): p. 4317-4320.
271. Zilberman, Y., et al., Nanoarray of polycyclic aromatic hydrocarbons and carbon
nanotubes for accurate and predictive detection in real-world environmental humidity.
ACS nano, 2011. 5(8): p. 6743-6753.
170
272. Peng, G., E. Trock, and H. Haick, Detecting simulated patterns of lung cancer biomarkers
by random network of single-walled carbon nanotubes coated with nonpolymeric
organic materials. Nano letters, 2008. 8(11): p. 3631-3635.
273. Zhang, Y., et al., Identification of volatile biomarkers of gastric cancer cells and
ultrasensitive electrochemical detection based on sensing interface of Au-Ag alloy coated
MWCNTs. Theranostics, 2014. 4(2): p. 154.
274. Marom, O., et al., Gold nanoparticle sensors for detecting chronic kidney disease and
disease progression. Nanomedicine, 2012. 7(5): p. 639-650.
275. Nakhleh, M.K., et al., Detecting active pulmonary tuberculosis with a breath test using
nanomaterial-based sensors. European Respiratory Journal, 2014. 43(5): p. 1522-1525.
276. Haick, H., et al., Sniffing chronic renal failure in rat model by an array of random
networks of single-walled carbon nanotubes. ACS nano, 2009. 3(5): p. 1258-1266.
277. Ionescu, R., et al., Detection of multiple sclerosis from exhaled breath using bilayers of
polycyclic aromatic hydrocarbons and single-wall carbon nanotubes. ACS Chemical
Neuroscience, 2011. 2(12): p. 687-693.
278. Gardner, J., et al., The prediction of bacteria type and culture growth phase by an
electronic nose with a multi-layer perceptron network. Measurement Science and
Technology, 1998. 9(1): p. 120.
279. Holmberg, M., et al., Bacteria classification based on feature extraction from sensor
data. Biotechnology techniques, 1998. 12(4): p. 319-324.
280. Pavlou, A.K., et al., An intelligent rapid odour recognition model in discrimination of
Helicobacter pylori and other gastroesophageal isolates in vitro. Biosensors and
Bioelectronics, 2000. 15(7-8): p. 333-342.
281. Lai, S.Y., et al., Identification of upper respiratory bacterial pathogens with the electronic
nose. The Laryngoscope, 2002. 112(6): p. 975-979.
282. Pavlou, A., A. Turner, and N. Magan, Recognition of anaerobic bacterial isolates in vitro
using electronic nose technology. Letters in applied microbiology, 2002. 35(5): p. 366-
369.
283. Pavlou, A.K., et al., Detection of Mycobacterium tuberculosis (TB) in vitro and in situ
using an electronic nose in combination with a neural network system. Biosensors and
Bioelectronics, 2004. 20(3): p. 538-544.
284. Lykos, P., et al., Rapid detection of bacteria from blood culture by an electronic nose.
Journal of Microbiology, 2001. 39(3): p. 213-218.
171
285. Gibson, T., et al., Detection and simultaneous identification of microorganisms from
headspace samples using an electronic nose. Sensors and Actuators B: Chemical, 1997.
44(1-3): p. 413-422.
286. Fend, R., et al., Prospects for clinical application of electronic-nose technology to early
detection of Mycobacterium tuberculosis in culture and sputum. Journal of Clinical
Microbiology, 2006. 44(6): p. 2039-2045.
287. Dutta, R., et al., Bacteria classification using Cyranose 320 electronic nose. Biomedical
engineering online, 2002. 1(1): p. 1-7.
288. Bruins, M., et al., Device-independent, real-time identification of bacterial pathogens
with a metal oxide-based olfactory sensor. European journal of clinical microbiology &
infectious diseases, 2009. 28(7): p. 775-780.
289. Wilson, A.D., D. Lester, and C. Oberle, Development of conductive polymer analysis for
the rapid detection and identification of phytopathogenic microbes. Phytopathology,
2004. 94(5): p. 419-431.
290. Siripatrawan, U., Rapid differentiation between E. coli and Salmonella typhimurium using
metal oxide sensors integrated with pattern recognition. Sensors and Actuators B:
Chemical, 2008. 133(2): p. 414-419.
291. Roine, A., et al., Rapid and accurate detection of urinary pathogens by mobile IMS-based
electronic nose: A proof-of-principle study. PLoS One, 2014. 9(12): p. e114279.
292. Parry, A., et al., Leg ulcer odour detection identifies β-haemolytic streptococcal infection.
Journal of wound care, 1995. 4(9): p. 404-406.
293. Shykhon, M., et al., Clinical evaluation of the electronic nose in the diagnosis of ear, nose
and throat infection: a preliminary study. The Journal of Laryngology & Otology, 2004.
118(9): p. 706-709.
294. Chandiok, S., et al., Screening for bacterial vaginosis: a novel application of artificial nose
technology. Journal of clinical pathology, 1997. 50(9): p. 790-791.
295. Hay, P., et al., Evaluation of a novel diagnostic test for bacterial vaginosis:'the electronic
nose'. International journal of STD & AIDS, 2003. 14(2): p. 114-118.
296. Chaudry, A.N., et al., Analysis of vaginal acetic acid in patients undergoing treatment for
bacterial vaginosis. Journal of clinical microbiology, 2004. 42(11): p. 5170-5175.
297. Persaud, K.C., et al., Monitoring urinary tract infections and bacterial vaginosis. Sensors
and Actuators B: Chemical, 2006. 116(1-2): p. 116-120.
172
298. Aathithan, S., et al., Diagnosis of bacteriuria by detection of volatile organic compounds
in urine using an automated headspace analyzer with multiple conducting polymer
sensors. Journal of clinical microbiology, 2001. 39(7): p. 2590-2593.
299. Pavlou, A.K., et al., Use of an electronic nose system for diagnoses of urinary tract
infections. Biosensors and Bioelectronics, 2002. 17(10): p. 893-899.
300. Kodogiannis, V. and E. Wadge, The use of gas-sensor arrays to diagnose urinary tract
infections. International journal of neural systems, 2005. 15(05): p. 363-376.
301. Arasaradnam, R., et al., Evaluation of gut bacterial populations using an electronic e-
nose and field asymmetric ion mobility spectrometry: further insights into
‘fermentonomics’. Journal of medical engineering & technology, 2012. 36(7): p. 333-337.
302. Kodogiannis, V.S., et al., Artificial odor discrimination system using electronic nose and
neural networks for the identification of urinary tract infection. IEEE Transactions on
information technology in biomedicine, 2008. 12(6): p. 707-713.
303. Yates, J.W., et al., Data reduction in headspace analysis of blood and urine samples for
robust bacterial identification. Computer methods and programs in biomedicine, 2005.
79(3): p. 259-271.
304. Capelli, L., et al., Application and uses of electronic noses for clinical diagnosis on urine
samples: A review. Sensors, 2016. 16(10): p. 1708.
305. Hanson, C. and H. Steinberger. The use of a novel'electronic nose'to diagnose the
presence of intrapulmonary infection. in Anesthesiology. 1997. LIPPINCOTT-RAVEN PUBL
227 EAST WASHINGTON SQ, PHILADELPHIA, PA 19106.
306. Hockstein, N.G., et al., Diagnosis of pneumonia with an electronic nose: correlation of
vapor signature with chest computed tomography scan findings. The Laryngoscope,
2004. 114(10): p. 1701-1705.
307. Hockstein, N.G., et al., Correlation of pneumonia score with electronic nose signature: A
prospective study. Annals of Otology, Rhinology & Laryngology, 2005. 114(7): p. 504-
508.
308. Hanson, C.W. and E.R. Thaler, Electronic nose prediction of a clinical pneumonia score:
biosensors and microbes. The Journal of the American Society of Anesthesiologists,
2005. 102(1): p. 63-68.
309. Mohamed, E.I., et al. A novel method for diagnosing chronic rhinosinusitis based on an
electronic nose. in Anales otorrinolaringológicos ibero-americanos. 2003.
173
310. Bruno, E., et al., Can the electronic nose diagnose chronic rhinosinusitis? A new
experimental study. European archives of oto-rhino-laryngology, 2008. 265(4): p. 425-
428.
311. Saha, D., Insect Olfaction in Chemical Sensing, in Canines. 2022, Jenny Stanford
Publishing. p. 151-177.
312. Röck, F., N. Barsan, and U. Weimar, Electronic nose: current status and future trends.
Chemical reviews, 2008. 108(2): p. 705-725.
313. Harper, W.J., The strengths and weaknesses of the electronic nose. Headspace analysis
of foods and flavors, 2001: p. 59-71.
314. Rudnitskaya, A., Calibration update and drift correction for electronic noses and tongues.
Frontiers in chemistry, 2018. 6: p. 433.
315. Scheepers, M.H., et al., Diagnostic Performance of Electronic Noses in Cancer Diagnoses
Using Exhaled Breath: A Systematic Review and Meta-analysis. JAMA Network Open,
2022. 5(6): p. e2219372-e2219372.
316. Göpel, W., From electronic to bioelectronic olfaction, or: from artificial “moses” to real
noses. Sensors and Actuators B: Chemical, 2000. 65(1-3): p. 70-72.
317. Gaillard, I., S. Rouquier, and D. Giorgi, Olfactory receptors. Cellular and Molecular Life
Sciences CMLS, 2004. 61(4): p. 456-469.
318. Berna, A.Z., A.R. Anderson, and S.C. Trowell, Bio-benchmarking of electronic nose
sensors. PloS one, 2009. 4(7): p. e6406.
319. Janata, J., Principles of chemical sensors. 2010: Springer Science & Business Media.
320. Safran, M., et al., Human gene-centric databases at the Weizmann Institute of Science:
GeneCards, UDB, CroW 21 and HORDE. Nucleic acids research, 2003. 31(1): p. 142-146.
321. Pham-Delegue, M., et al., Sunflower volatiles involved in honeybee discrimination among
genotypes and flowering stages. Journal of Chemical Ecology, 1989. 15(1): p. 329-343.
322. Dobson, H.E., Floral volatiles in insect biology. Insect-plant interactions, 2017: p. 47-82.
323. Sandoz, J.-C., Olfaction in honey bees: from molecules to behavior. Honeybee
neurobiology and behavior, 2012: p. 235-252.
324. Bruce, T.J. and J.A. Pickett, Perception of plant volatile blends by herbivorous insects–
finding the right mix. Phytochemistry, 2011. 72(13): p. 1605-1611.
174
325. Ache, B.W. and J.M. Young, Olfaction: diverse species, conserved principles. Neuron,
2005. 48(3): p. 417-430.
326. Rouquier, S., A. Blancher, and D. Giorgi, The olfactory receptor gene repertoire in
primates and mouse: evidence for reduction of the functional fraction in primates.
Proceedings of the National Academy of Sciences, 2000. 97(6): p. 2870-2874.
327. Malnic, B., P.A. Godfrey, and L.B. Buck, The human olfactory receptor gene family.
Proceedings of the National Academy of Sciences, 2004. 101(8): p. 2584-2589.
328. Bushdid, C., et al., Humans can discriminate more than 1 trillion olfactory stimuli.
Science, 2014. 343(6177): p. 1370-1372.
329. Niimura, Y. and M. Nei, Evolution of olfactory receptor genes in the human genome.
Proceedings of the National Academy of Sciences, 2003. 100(21): p. 12235-12240.
330. Ressler, K.J., S.L. Sullivan, and L.B. Buck, Information coding in the olfactory system:
evidence for a stereotyped and highly organized epitope map in the olfactory bulb. Cell,
1994. 79(7): p. 1245-1255.
331. Vassar, R., et al., Topographic organization of sensory projections to the olfactory bulb.
Cell, 1994. 79(6): p. 981-991.
332. Balu, R., R.T. Pressler, and B.W. Strowbridge, Multiple modes of synaptic excitation of
olfactory bulb granule cells. Journal of Neuroscience, 2007. 27(21): p. 5621-5632.
333. Buck, L.B. and C. Bargmann, Smell and taste: The chemical senses. Principles of neural
science, 2000. 4: p. 625-647.
334. Desmaisons, D., J.-D. Vincent, and P.-M. Lledo, Control of action potential timing by
intrinsic subthreshold oscillations in olfactory bulb output neurons. Journal of
Neuroscience, 1999. 19(24): p. 10727-10737.
335. Schoppa, N.E. and G.L. Westbrook, Glomerulus-specific synchronization of mitral cells in
the olfactory bulb. Neuron, 2001. 31(4): p. 639-651.
336. Galán, R.F., et al., Correlation-induced synchronization of oscillations in olfactory bulb
neurons. Journal of Neuroscience, 2006. 26(14): p. 3646-3655.
337. Imamura, F., A. Ito, and B.J. LaFever, Subpopulations of projection neurons in the
olfactory bulb. Frontiers in Neural Circuits, 2020. 14: p. 561822.
338. Gilad, Y., et al., Loss of olfactory receptor genes coincides with the acquisition of full
trichromatic vision in primates. PLoS biology, 2004. 2(1): p. e5.
175
339. Elgar, M.A., et al., Focus: ecology and evolution: insect antennal morphology: the
evolution of diverse solutions to odorant perception. The Yale journal of biology and
medicine, 2018. 91(4): p. 457.
340. Carraher, C., et al., Towards an understanding of the structural basis for insect olfaction
by odorant receptors. Insect Biochemistry and Molecular Biology, 2015. 66: p. 31-41.
341. Carey, A.F. and J.R. Carlson, Insect olfaction from model systems to disease control.
Proceedings of the National Academy of Sciences, 2011. 108(32): p. 12987-12995.
342. Ochieng, S.A., E. Hallberg, and B. Hansson, Fine structure and distribution of antennal
sensilla of the desert locust, Schistocerca gregaria (Orthoptera: Acrididae). Cell and
tissue research, 1998. 291(3): p. 525-536.
343. Hallem, E.A. and J.R. Carlson, Coding of odors by a receptor repertoire. Cell, 2006.
125(1): p. 143-160.
344. Masson, C. and H. Mustaparta, Chemical information processing in the olfactory system
of insects. Physiological Reviews, 1990. 70(1): p. 199-245.
345. Stocker, R.F., The organization of the chemosensory system in Drosophila melanogaster:
a rewiew. Cell and tissue research, 1994. 275: p. 3-26.
346. Sato, K., et al., Insect olfactory receptors are heteromeric ligand-gated ion channels.
Nature, 2008. 452(7190): p. 1002-1006.
347. Benton, R., et al., Atypical membrane topology and heteromeric function of Drosophila
odorant receptors in vivo. PLoS biology, 2006. 4(2): p. e20.
348. Buck, L. and R. Axel, A novel multigene family may encode odorant receptors: a
molecular basis for odor recognition. Cell, 1991. 65(1): p. 175-187.
349. Egea-Weiss, A., et al., High precision of spike timing across olfactory receptor neurons
allows rapid odor coding in Drosophila. IScience, 2018. 4: p. 76-83.
350. Wright, G.A., M. Carlton, and B.H. Smith, A honeybee's ability to learn, recognize, and
discriminate odors depends upon odor sampling time and concentration. Behavioral
neuroscience, 2009. 123(1): p. 36.
351. Krofczik, S., R. Menzel, and M.P. Nawrot, Rapid odor processing in the honeybee
antennal lobe network. Frontiers in computational neuroscience, 2009: p. 9.
352. Wicher, D., Olfactory signaling in insects. Progress in molecular biology and translational
science, 2015. 130: p. 37-54.
176
353. Vosshall, L.B., A.M. Wong, and R. Axel, An olfactory sensory map in the fly brain. Cell,
2000. 102(2): p. 147-159.
354. Couto, A., M. Alenius, and B.J. Dickson, Molecular, anatomical, and functional
organization of the Drosophila olfactory system. Current Biology, 2005. 15(17): p. 1535-
1547.
355. Hansson, B.S. and M.C. Stensmyr, Evolution of insect olfaction. Neuron, 2011. 72(5): p.
698-711.
356. Hansson, B.S., et al., Functional specialization of olfactory glomeruli in a moth. Science,
1992. 256(5061): p. 1313-1315.
357. Kelber, C., W. Rössler, and C.J. Kleineidam, Multiple olfactory receptor neurons and their
axonal projections in the antennal lobe of the honeybee Apis mellifera. Journal of
Comparative Neurology, 2006. 496(3): p. 395-405.
358. Brockmann, A. and D. Brückner, Projection pattern of poreplate sensory neurones in
honey bee worker, Apis mellifera L.(Hymenoptera: Apidae). International Journal of
Insect Morphology and Embryology, 1995. 24(4): p. 405-411.
359. Laurent, G. and M. Naraghi, Odorant-induced oscillations in the mushroom bodies of the
locust. Journal of Neuroscience, 1994. 14(5): p. 2993-3004.
360. Perez-Orive, J., et al., Oscillations and sparsening of odor representations in the
mushroom body. Science, 2002. 297(5580): p. 359-365.
361. Kay, L.M. and M. Stopfer. Information processing in the olfactory systems of insects and
vertebrates. in Seminars in cell & developmental biology. 2006. Elsevier.
362. Bazhenov, M., et al., Fast odor learning improves reliability of odor responses in the
locust antennal lobe. Neuron, 2005. 46(3): p. 483-492.
363. Hosler, J.S., K.L. Buxton, and B.H. Smith, Impairment of olfactory discrimination by
blockade of GABA and nitric oxide activity in the honey bee antennal lobes. Behavioral
neuroscience, 2000. 114(3): p. 514.
364. Stopfer, M., et al., Impaired odour discrimination on desynchronization of odour-
encoding neural assemblies. Nature, 1997. 390(6655): p. 70-74.
365. Szyszka, P., et al., Sparsening and temporal sharpening of olfactory representations in
the honeybee mushroom bodies. Journal of neurophysiology, 2005. 94(5): p. 3303-3313.
366. Wang, Y., et al., Stereotyped odor-evoked activity in the mushroom body of Drosophila
revealed by green fluorescent protein-based Ca2+ imaging. Journal of Neuroscience,
2004. 24(29): p. 6507-6514.
177
367. Szyszka, P., A. Galkin, and R. Menzel, Associative and non-associative plasticity in Kenyon
cells of the honeybee mushroom body. Frontiers in systems neuroscience, 2008: p. 3.
368. Boeckh, J. and L.P. Tolbert, Synaptic organization and development of the antennal lobe
in insects. Microscopy research and technique, 1993. 24(3): p. 260-280.
369. Angioy, A.M., et al., Extreme sensitivity in an olfactory system. Chemical Senses, 2003.
28(4): p. 279-284.
370. Kaissling, K.-E. and E. Priesner, Die riechschwelle des seidenspinners.
Naturwissenschaften, 1970. 57(1): p. 23-28.
371. Nei, M., Y. Niimura, and M. Nozawa, The evolution of animal chemosensory receptor
gene repertoires: roles of chance and necessity. Nature Reviews Genetics, 2008. 9(12): p.
951-963.
372. Pinching, A. and T. Powell, The neuropil of the glomeruli of the olfactory bulb. Journal of
cell science, 1971. 9(2): p. 347-377.
373. Gazit, I. and J. Terkel, Explosives detection by sniffer dogs following strenuous physical
activity. Applied Animal Behaviour Science, 2003. 81(2): p. 149-161.
374. Caygill, J.S., F. Davis, and S.P. Higson, Current trends in explosive detection techniques.
Talanta, 2012. 88: p. 14-29.
375. Browne, C., K. Stafford, and R. Fordham, The use of scent-detection dogs. Irish
Veterinary Journal, 2006. 59(2): p. 97.
376. Habib, M.K., Controlled biological and biomimetic systems for landmine detection.
Biosensors and Bioelectronics, 2007. 23(1): p. 1-18.
377. Lazarowski, L. and D.C. Dorman, Explosives detection by military working dogs: Olfactory
generalization from components to mixtures. Applied Animal Behaviour Science, 2014.
151: p. 84-93.
378. Lubow, R.E., The war animals. 1977: Doubleday Books.
379. Brown, S.W. and L.H. Goldstein, Can seizure-alert dogs predict seizures? Epilepsy
research, 2011. 97(3): p. 236-242.
380. Catala, A., et al., Dogs demonstrate the existence of an epileptic seizure odour in
humans. Scientific reports, 2019. 9(1): p. 1-7.
381. Wells, D.L., S.W. Lawson, and A.N. Siriwardena, Canine responses to hypoglycemia in
patients with type 1 diabetes. The Journal of Alternative and Complementary Medicine,
2008. 14(10): p. 1235-1241.
178
382. Weber, K.S., M. Roden, and K. Müssig, Do dogs sense hypoglycaemia? Diabetic
Medicine, 2016. 33(7): p. 934-938.
383. Reeve, C., et al., An idiographic investigation of diabetic alert dogs' ability to learn from
a small sample of breath samples from people with Type 1 diabetes. Canadian journal of
diabetes, 2020. 44(1): p. 37-43. e1.
384. Guest, C., et al., Trained dogs identify people with malaria parasites by their odour. The
Lancet Infectious Diseases, 2019. 19(6): p. 578-580.
385. Kokocińska-Kusiak, A., et al., Canine olfaction: physiology, behavior, and possibilities for
practical applications. Animals, 2021. 11(8): p. 2463.
386. Angle, T.C., et al., Real-time detection of a virus using detection dogs. Frontiers in
veterinary science, 2016: p. 79.
387. Williams, H. and A. Pembroke, Sniffer dogs in the melanoma clinic? The Lancet, 1989.
333(8640): p. 734.
388. Church, J. and H. Williams, Another sniffer dog for the clinic? The Lancet, 2001.
358(9285): p. 930.
389. Fraser, L., Scientists put sniffer dogs on the scent of men with cancer. Sunday Telegraph,
2002. 2.
390. Dobson, R., Dogs can sniff out first signs of men’s cancer. Sunday Times, 2003. 27(5).
391. McCulloch, M., et al., Diagnostic accuracy of canine scent detection in early-and late-
stage lung and breast cancers. Integrative cancer therapies, 2006. 5(1): p. 30-39.
392. Ehmann, R., et al., Canine scent detection in the diagnosis of lung cancer: revisiting a
puzzling phenomenon. European respiratory journal, 2012. 39(3): p. 669-676.
393. Amundsen, T., et al., Can dogs smell lung cancer? First study using exhaled breath and
urine screening in unselected patients with suspected lung cancer. Acta oncologica,
2014. 53(3): p. 307-315.
394. Pickel, D., et al., Evidence for canine olfactory detection of melanoma. Applied Animal
Behaviour Science, 2004. 89(1-2): p. 107-116.
395. Cornu, J.-N., et al., Olfactory detection of prostate cancer by dogs sniffing urine: a step
forward in early diagnosis. European urology, 2011. 59(2): p. 197-201.
396. Sonoda, H., et al., Colorectal cancer screening with odour material by canine scent
detection. Gut, 2011. 60(6): p. 814-819.
179
397. Horvath, G., et al., Human ovarian carcinomas detected by specific odor. Integrative
cancer therapies, 2008. 7(2): p. 76-80.
398. Elliker, K.R., et al., Key considerations for the experimental training and evaluation of
cancer odour detection dogs: lessons learnt from a double-blind, controlled trial of
prostate cancer detection. BMC urology, 2014. 14(1): p. 1-9.
399. Jezierski, T., et al., Study of the art: canine olfaction used for cancer detection on the
basis of breath odour. Perspectives and limitations. Journal of breath research, 2015.
9(2): p. 027001.
400. Willis, C.M., et al., Olfactory detection of human bladder cancer by dogs: proof of
principle study. Bmj, 2004. 329(7468): p. 712.
401. Gordon, R.T., et al., The use of canines in the detection of human cancers. The Journal of
Alternative and Complementary Medicine, 2008. 14(1): p. 61-67.
402. Taverna, G., et al., Olfactory system of highly trained dogs detects prostate cancer in
urine samples. The Journal of urology, 2015. 193(4): p. 1382-1387.
403. Taverna, G., et al., Highly-trained dogs’ olfactory system for detecting biochemical
recurrence following radical prostatectomy. Clinical Chemistry and Laboratory Medicine
(CCLM), 2016. 54(3): p. e67-e70.
404. Moser, E. and M. McCulloch, Canine scent detection of human cancers: A review of
methods and accuracy. Journal of Veterinary Behavior, 2010. 5(3): p. 145-152.
405. Boedeker, E., G. Friedel, and T. Walles, Sniffer dogs as part of a bimodal bionic research
approach to develop a lung cancer screening. Interactive cardiovascular and thoracic
surgery, 2012. 14(5): p. 511-515.
406. Lippi, G. and G. Cervellin, Canine olfactory detection of cancer versus laboratory testing:
myth or opportunity? Clinical chemistry and laboratory medicine, 2012. 50(3): p. 435-
439.
407. Manzini, I., D. Schild, and C. Di Natale, Principles of odor coding in vertebrates and
artificial chemosensory systems. Physiological Reviews, 2022. 102(1): p. 61-154.
408. Angle, C., et al., Canine detection of the volatilome: a review of implications for
pathogen and disease detection. Frontiers in veterinary science, 2016. 3: p. 47.
409. Concha, A.R., et al., Canine olfactory thresholds to amyl acetate in a biomedical
detection scenario. Frontiers in Veterinary Science, 2019. 5: p. 345.
410. Walker, D.B., et al., Naturalistic quantification of canine olfactory sensitivity. Applied
Animal Behaviour Science, 2006. 97(2-4): p. 241-254.
180
411. Craven, B.A., E.G. Paterson, and G.S. Settles, The fluid dynamics of canine olfaction:
unique nasal airflow patterns as an explanation of macrosmia. Journal of the Royal
Society Interface, 2010. 7(47): p. 933-943.
412. Jenkins, E.K., M.T. DeChant, and E.B. Perry, When the nose doesn’t know: Canine
olfactory function associated with health, management, and potential links to
microbiota. Frontiers in veterinary science, 2018: p. 56.
413. Walczak, M., et al., Impact of individual training parameters and manner of taking
breath odor samples on the reliability of canines as cancer screeners. Journal of
veterinary behavior, 2012. 7(5): p. 283-294.
414. Edwards, T.L., et al., Animal olfactory detection of human diseases: Guidelines and
systematic review. Journal of veterinary behavior, 2017. 20: p. 59-73.
415. Weetjens, B., et al., African pouched rats for the detection of pulmonary tuberculosis in
sputum samples. The International journal of tuberculosis and lung disease, 2009. 13(6):
p. 737-743.
416. Poling, A., et al., Tuberculosis detection by giant African pouched rats. The Behavior
Analyst, 2011. 34(1): p. 47-54.
417. Poling, A., et al., Using giant African pouched rats to detect tuberculosis in human
sputum samples: 2009 findings. The American journal of tropical medicine and hygiene,
2010. 83(6): p. 1308.
418. Mgode, G.F., et al., Diagnosis of tuberculosis by trained African giant pouched rats and
confounding impact of pathogens and microflora of the respiratory tract. Journal of
clinical microbiology, 2012. 50(2): p. 274-280.
419. Mgode, G.F., et al., Mycobacterium tuberculosis volatiles for diagnosis of tuberculosis by
Cricetomys rats. Tuberculosis, 2012. 92(6): p. 535-542.
420. Rospars, J.-P., et al., Heterogeneity and convergence of olfactory first-order neurons
account for the high speed and sensitivity of second-order neurons. PLoS computational
biology, 2014. 10(12): p. e1003975.
421. Liao, C., et al., Behavioural and genetic evidence for C. elegans' ability to detect volatile
chemicals associated with explosives. PloS one, 2010. 5(9): p. e12615.
422. Neto, M.F., et al., The nematode Caenorhabditis elegans displays a chemotaxis behavior
to tuberculosis-specific odorants. Journal of clinical tuberculosis and other mycobacterial
diseases, 2016. 4: p. 44-49.
423. Hirotsu, T., et al., A highly accurate inclusive cancer screening test using Caenorhabditis
elegans scent detection. PloS one, 2015. 10(3): p. e0118699.
181
424. Ueda, Y., et al., Application of C. elegans cancer screening test for the detection of
pancreatic tumor in genetically engineered mice. Oncotarget, 2019. 10(52): p. 5412.
425. Kusumoto, H., et al., Efficiency of gastrointestinal cancer detection by Nematode-NOSE
(N-NOSE). in vivo, 2020. 34(1): p. 73-80.
426. Lanza, E., et al., C. elegans-based chemosensation strategy for the early detection of
cancer metabolites in urine samples. Scientific reports, 2021. 11(1): p. 1-16.
427. Thompson, M., et al., A Caenorhabditis elegans behavioral assay distinguishes early
stage prostate cancer patient urine from controls. Biology open, 2021. 10(3): p.
bio057398.
428. di Luccio, E., M. Morishita, and T. Hirotsu, C. elegans as a Powerful Tool for Cancer
Screening. Biomedicines, 2022. 10(10): p. 2371.
429. Olson, D., et al., Parasitic wasps learn and report diverse chemicals with unique
conditionable behaviors. Chemical senses, 2003. 28(6): p. 545-549.
430. Tomberlin, J.K., et al., Conditioned Microplitis croceipes Cresson (Hymenoptera:
Braconidae) detect and respond to 2, 4-DNT: development of a biological sensor. Journal
of Forensic Science, 2005. 50(5): p. JFS2005014-4.
431. Rains, G.C., S.L. Utley, and W.J. Lewis, Behavioral monitoring of trained insects for
chemical detection. Biotechnology progress, 2006. 22(1): p. 2-8.
432. Rains, G.C., et al., Limits of volatile chemical detection of a parasitoid wasp, Microplitis
croceipes, and an electronic nose: a comparative study. Transactions of the ASAE, 2004.
47(6): p. 2145.
433. Meiners, T., F. Wäckers, and W.J. Lewis, The effect of molecular structure on olfactory
discrimination by the parasitoid Microplitis croceipes. Chemical senses, 2002. 27(9): p.
811-816.
434. MacDonald, J., et al., Alternatives for landmine detection. 2003, RAND CORP SANTA
MONICA CA.
435. Repasky, K.S., et al., Optical detection of honeybees by use of wing-beat modulation of
scattered laser light for locating explosives and land mines. Applied optics, 2006. 45(8):
p. 1839-1843.
436. Taylor-mccabe, K.J., R.M. Wingo, and T.K. Haarmann, Honey bees (Apis mellifera) as
explosives detectors: exploring proboscis extension reflex conditioned response to
trinitrotolulene (TNT). 2008, Los Alamos National Lab.(LANL), Los Alamos, NM (United
States).
182
437. Leitch, O., et al., Biological organisms as volatile compound detectors: A review. Forensic
science international, 2013. 232(1-3): p. 92-103.
438. Suckling, D.M. and R.L. Sagar, Honeybees Apis mellifera can detect the scent of
Mycobacterium tuberculosis. Tuberculosis, 2011. 91(4): p. 327-328.
439. Davis, N., How honey bees may one day help detect signs of cancers, in The Guardian.
2013.
440. Bryant, R. Honey bees can be trained to detect cancer "in ten minutes" says designer.
2013 2022, November 20]; Available from:
https://www.dezeen.com/2013/11/20/honey-bees-can-be-trained-to-detect-cancer-in-
ten-minutes-says-designer-susana-soares/.
441. King, T.L., et al., Explosives detection with hard-wired moths. IEEE Transactions on
Instrumentation and Measurement, 2004. 53(4): p. 1113-1118.
442. Rains, G.C., J.K. Tomberlin, and D. Kulasiri, Using insect sniffing devices for detection.
Trends in biotechnology, 2008. 26(6): p. 288-294.
443. Roelofs, W.L., Electroantennogram assays: rapid and convenient screening procedures
for pheromones, in Techniques in pheromone research. 1984, Springer. p. 131-159.
444. Schott, M., et al., Insect antenna-based biosensors for in situ detection of volatiles.
Yellow Biotechnology II, 2013: p. 101-122.
445. Guo, M., et al., Odorant receptors for detecting flowering plant cues are functionally
conserved across moths and butterflies. Molecular biology and evolution, 2021. 38(4): p.
1413-1427.
446. Färbert, P., et al., Measuring pheromone concentrations in cotton fields with the EAG
method, in Insect Pheromone Research. 1997, Springer. p. 347-358.
447. Van der Pers, J. and A. Minks, A portable electroantennogram sensor for routine
measurements of pheromone concentrations in greenhouses. Entomologia
experimentalis et applicata, 1998. 87(2): p. 209-215.
448. Sauer, A.E., et al., A portable EAG system for the measurement of pheromone
concentrations in the field. Chemical Senses, 1992. 17(5): p. 543-553.
449. Schiestl, F. and F. Marion-Poll, Detection of physiologically active flower volatiles using
gas chromatography coupled with electroantennography, in Analysis of taste and
aroma. 2002, Springer. p. 173-198.
183
450. Moore, I., Biological amplification for increasing electroantennogram discrimination
between two female sex pheromones ofSpodoptera littoralis (Lepidoptera: Noctuidae).
Journal of Chemical Ecology, 1981. 7(5): p. 791-798.
451. Park, K.C. and T.C. Baker, Improvement of signal-to-noise ratio in electroantennogram
responses using multiple insect antennae. Journal of Insect Physiology, 2002. 48(12): p.
1139-1145.
452. Park, K.C., et al., Odor discrimination using insect electroantennogram responses from
an insect antennal array. Chemical senses, 2002. 27(4): p. 343-352.
453. Cork, A., et al., Gas chromatography linked to electroantennography: a versatile
technique for identifying insect semiochemicals, in Chromatography and isolation of
insect hormones and pheromones. 1990, Springer. p. 271-279.
454. Moorhouse, J., et al., Method for use in studies of insect chemical communication.
Nature, 1969. 223(5211): p. 1174-1175.
455. Struble, D.L. and H. Arn, Combined gas chromatography and electroantennogram
recording of insect olfactory responses, in Techniques in pheromone research. 1984,
Springer. p. 161-178.
456. Sillam-Dusses, D., et al., Identification by GC-EAD of the two-component trail-following
pheromone of Prorhinotermes simplex (Isoptera, Rhinotermitidae, Prorhinotermitinae).
Journal of Insect Physiology, 2009. 55(8): p. 751-757.
457. Kurtovic, A., A. Widmer, and B.J. Dickson, A single class of olfactory neurons mediates
behavioural responses to a Drosophila sex pheromone. Nature, 2007. 446(7135): p. 542-
546.
458. Siderhurst, M.S. and E.B. Jang, Female-biased attraction of oriental fruit fly, Bactrocera
dorsalis (Hendel), to a blend of host fruit volatiles from Terminalia catappa L. Journal of
chemical ecology, 2006. 32(11): p. 2513-2524.
459. Olsson, P.-O.C., et al., Electrophysiological and behavioral responses to chocolate
volatiles in both sexes of the pyralid moths Ephestia cautella and Plodia interpunctella.
Journal of chemical ecology, 2005. 31(12): p. 2947-2961.
460. Zhang, Q.-H., N. Erbilgin, and S.J. Seybold, GC-EAD responses to semiochemicals by eight
beetles in the subcortical community associated with Monterey pine trees in coastal
California: similarities and disparities across three trophic levels. Chemoecology, 2008.
18(4): p. 243-254.
461. Van Der Pers, J., Comparison of electroantennogram response spectra to plant volatiles
in seven species of Yponomeuta and in the tortricid Adoxophyes orana. Entomologia
experimentalis et applicata, 1981. 30(2): p. 181-192.
184
462. Park, K.C., et al., Electroantennogram responses of a parasitic wasp, Microplitis
croceipes, to host-related volatile and anthropogenic compounds. Physiological
Entomology, 2001. 26(1): p. 69-77.
463. Revadi, S., et al., Olfactory responses of D rosophila suzukii females to host plant
volatiles. Physiological Entomology, 2015. 40(1): p. 54-64.
464. Schroth, P., et al., Extending the capabilities of an antenna/chip biosensor by employing
various insect species. Sensors and Actuators B: Chemical, 2001. 78(1-3): p. 1-5.
465. Schöning, M.J., et al., A BioFET on the basis of intact insect antennae. Sensors and
Actuators B: Chemical, 1998. 47(1-3): p. 235-238.
466. Schroth, P., et al., Insect-based BioFETs with improved signal characteristics. Biosensors
and Bioelectronics, 1999. 14(3): p. 303-308.
467. Schütz, S., et al., An insect-based BioFET as a bioelectronic nose. Sensors and Actuators
B: Chemical, 2000. 65(1-3): p. 291-295.
468. Schöning, M.J., P. Schroth, and S. Schütz, The use of insect chemoreceptors for the
assembly of biosensors based on semiconductor field-effect transistors. Electroanalysis:
An International Journal Devoted to Fundamental and Practical Aspects of
Electroanalysis, 2000. 12(9): p. 645-652.
469. Wadhams, L.J., Coupled gas chromatography—single cell recording: a new technique for
use in the analysis of insect pheromones. Zeitschrift für Naturforschung C, 1982. 37(10):
p. 947-952.
470. Wadhams, L., The use of coupled gas chromatography: electrophysiological techniques
in the identification of insect pheromones, in Chromatography and isolation of insect
hormones and pheromones. 1990, Springer. p. 289-298.
471. Blight, M., et al., Antennal perception of oilseed rape, Brassica napus (Brassicaceae),
volatiles by the cabbage seed weevilCeutorhynchus assimilis (Coleoptera, Curculionidae).
Journal of chemical ecology, 1995. 21(11): p. 1649-1664.
472. Hansson, B., et al., Correlation between dendrite diameter and action potential
amplitude in sex pheromone specific receptor neurons in male Ostrinia nubilalis
(Lepidoptera: Pyralidae). Tissue and Cell, 1994. 26(4): p. 503-512.
473. Olsson, S.B. and B.S. Hansson, Electroantennogram and single sensillum recording in
insect antennae, in Pheromone signaling. 2013, Springer. p. 157-177.
474. De Bruyne, M., K. Foster, and J.R. Carlson, Odor coding in the Drosophila antenna.
Neuron, 2001. 30(2): p. 537-552.
185
475. Hallem, E.A., M.G. Ho, and J.R. Carlson, The molecular basis of odor coding in the
Drosophila antenna. Cell, 2004. 117(7): p. 965-979.
476. Dobritsa, A.A., et al., Integrating the molecular and cellular basis of odor coding in the
Drosophila antenna. Neuron, 2003. 37(5): p. 827-841.
477. Mustaparta, H., Responses of single olfactory cells in the pine weevilHylobius abietis
L.(Col.: Curculionidae). Journal of Comparative Physiology, 1975. 97(4): p. 271-290.
478. Ma, W.C. and J. Visser, Single unit analysis of odour quality coding by the olfactory
antennal receptor system of the Colorado beetle. Entomologia Experimentalis et
Applicata, 1978. 24(3): p. 520-533.
479. Sass, H., Olfactory receptors on the antenna ofPeriplaneta: Response constellations that
encode food odors. Journal of comparative physiology, 1978. 128(3): p. 227-233.
480. Hansson, B., et al., Physiological responses and central nervous projections of antennal
olfactory receptor neurons in the adult desert locust, Schistocerca gregaria (Orthoptera:
Acrididae). Journal of Comparative Physiology A, 1996. 179(2): p. 157-167.
481. Ignell, R., S. Anton, and B. Hansson, Central nervous processing of behaviourally relevant
odours in solitary and gregarious fifth instar locusts, Schistocerca gregaria. Journal of
Comparative Physiology A, 1998. 183(4): p. 453-465.
482. MacLeod, K., A. Bäcker, and G. Laurent, Who reads temporal information contained
across synchronized and oscillatory spike trains? Nature, 1998. 395(6703): p. 693-698.
483. Marshall, B., C.G. Warr, and M. De Bruyne, Detection of volatile indicators of illicit
substances by the olfactory receptors of Drosophila melanogaster. Chemical senses,
2010. 35(7): p. 613-625.
484. Nowotny, T., et al., Drosophila olfactory receptors as classifiers for volatiles from
disparate real world applications. Bioinspiration & Biomimetics, 2014. 9(4): p. 046007.
485. Pelz, D., et al., The molecular receptive range of an olfactory receptor in vivo (Drosophila
melanogaster Or22a). Journal of neurobiology, 2006. 66(14): p. 1544-1563.
486. Silbering, A.F., et al., Complementary function and integrated wiring of the evolutionarily
distinct Drosophila olfactory subsystems. Journal of Neuroscience, 2011. 31(38): p.
13357-13375.
487. Cui, X., C. Wu, and L. Zhang, Electrophysiological response patterns of 16 olfactory
neurons from the trichoid sensilla to odorant from fecal volatiles in the locust, Locusta
migratoria manilensis. Archives of insect biochemistry and physiology, 2011. 77(2): p.
45-57.
186
488. Strauch, M., et al., More than apples and oranges-Detecting cancer with a fruit fly's
antenna. Scientific reports, 2014. 4(1): p. 1-9.
489. Bhandawat, V., et al., Sensory processing in the Drosophila antennal lobe increases
reliability and separability of ensemble odor representations. Nature neuroscience,
2007. 10(11): p. 1474-1482.
490. Olsen, S.R. and R.I. Wilson, Lateral presynaptic inhibition mediates gain control in an
olfactory circuit. Nature, 2008. 452(7190): p. 956-960.
491. Wilson, R.I., G.C. Turner, and G. Laurent, Transformation of olfactory representations in
the Drosophila antennal lobe. Science, 2004. 303(5656): p. 366-370.
492. Olsen, S.R., V. Bhandawat, and R.I. Wilson, Divisive normalization in olfactory population
codes. Neuron, 2010. 66(2): p. 287-299.
493. Silbering, A.F. and C.G. Galizia, Processing of odor mixtures in the Drosophila antennal
lobe reveals both global inhibition and glomerulus-specific interactions. Journal of
Neuroscience, 2007. 27(44): p. 11966-11977.
494. Kazama, H. and R.I. Wilson, Origins of correlated activity in an olfactory circuit. Nature
neuroscience, 2009. 12(9): p. 1136-1144.
495. Ng, M., et al., Transmission of olfactory information between three populations of
neurons in the antennal lobe of the fly. Neuron, 2002. 36(3): p. 463-474.
496. Hansson, B.S. and S. Anton, Function and morphology of the antennal lobe: new
developments. Annual review of entomology, 2000. 45(1): p. 203-231.
497. Moreaux, L.C. and G. Laurent, Estimating firing rates from calcium signals in locust
projection neurons in vivo. Frontiers in neural circuits, 2007: p. 2.
498. Sachse, S. and C.G. Galizia, Role of inhibition for temporal and spatial odor
representation in olfactory output neurons: a calcium imaging study. Journal of
neurophysiology, 2002. 87(2): p. 1106-1117.
499. Joerges, J., et al., Representations of odours and odour mixtures visualized in the
honeybee brain. Nature, 1997. 387(6630): p. 285-288.
500. Galizia, C.G., et al., The glomerular code for odor representation is species specific in the
honeybee Apis mellifera. Nature neuroscience, 1999. 2(5): p. 473-478.
501. Sachse, S., A. Rappert, and C.G. Galizia, The spatial representation of chemical structures
in the antennal lobe of honeybees: steps towards the olfactory code. European Journal
of Neuroscience, 1999. 11(11): p. 3970-3982.
187
502. Galizia, C.G., et al., Odour coding is bilaterally symmetrical in the antennal lobes of
honeybees (Apis mellifera). European Journal of Neuroscience, 1998. 10(9): p. 2964-
2974.
503. Galizia, C.G., et al., A semi-in-vivo preparation for optical recording of the insect brain.
Journal of neuroscience methods, 1997. 76(1): p. 61-69.
504. Mohamed, A.A., et al., Odor mixtures of opposing valence unveil inter-glomerular
crosstalk in the Drosophila antennal lobe. Nature Communications, 2019. 10(1): p. 1-17.
505. Ai, M., et al., Acid sensing by the Drosophila olfactory system. Nature, 2010. 468(7324):
p. 691-695.
506. Wang, J.W., et al., Two-photon calcium imaging reveals an odor-evoked map of activity
in the fly brain. Cell, 2003. 112(2): p. 271-282.
507. Carlsson, M.A., et al., Spatio-temporal Ca2+ dynamics of moth olfactory projection
neurones. European Journal of Neuroscience, 2005. 22(3): p. 647-657.
508. Moreaux, L.C., et al., Integrated neurophotonics: toward dense volumetric interrogation
of brain circuit activity—at depth and in real time. Neuron, 2020. 108(1): p. 66-92.
509. Ali, F. and A.C. Kwan, Interpreting in vivo calcium signals from neuronal cell bodies,
axons, and dendrites: a review. Neurophotonics, 2019. 7(1): p. 011402.
510. Grewe, B.F. and F. Helmchen, Optical probing of neuronal ensemble activity. Current
opinion in neurobiology, 2009. 19(5): p. 520-529.
511. Laurent, G. and H. Davidowitz, Encoding of olfactory information with oscillating neural
assemblies. Science, 1994. 265(5180): p. 1872-1875.
512. MacLeod, K. and G. Laurent, Distinct mechanisms for synchronization and temporal
patterning of odor-encoding neural assemblies. Science, 1996. 274(5289): p. 976-979.
513. Wehr, M. and G. Laurent, Odour encoding by temporal sequences of firing in oscillating
neural assemblies. Nature, 1996. 384(6605): p. 162-166.
514. Bazhenov, M., et al., Model of cellular and network mechanisms for odor-evoked
temporal patterning in the locust antennal lobe. Neuron, 2001. 30(2): p. 569-581.
515. Anton, S., R. Ignell, and B.S. Hansson, Developmental changes in the structure and
function of the central olfactory system in gregarious and solitary desert locusts.
Microscopy research and technique, 2002. 56(4): p. 281-291.
516. Stopfer, M., V. Jayaraman, and G. Laurent, Intensity versus identity coding in an olfactory
system. Neuron, 2003. 39(6): p. 991-1004.
188
517. Brown, S.L., J. Joseph, and M. Stopfer, Encoding a temporally structured stimulus with a
temporally structured neural representation. Nature Neuroscience, 2005. 8(11): p. 1568-
1576.
518. Mazor, O. and G. Laurent, Transient dynamics versus fixed points in odor representations
by locust antennal lobe projection neurons. Neuron, 2005. 48(4): p. 661-673.
519. Ito, I., et al., Sparse odor representation and olfactory learning. Nature neuroscience,
2008. 11(10): p. 1177-1184.
520. Daly, K.C., G.A. Wright, and B.H. Smith, Molecular features of odorants systematically
influence slow temporal responses across clusters of coordinated antennal lobe units in
the moth Manduca sexta. Journal of neurophysiology, 2004. 92(1): p. 236-254.
521. Wilson, R.I. and G. Laurent, Role of GABAergic inhibition in shaping odor-evoked
spatiotemporal patterns in the Drosophila antennal lobe. Journal of Neuroscience, 2005.
25(40): p. 9069-9079.
522. Masse, N.Y., G.C. Turner, and G.S. Jefferis, Olfactory information processing in
Drosophila. Current Biology, 2009. 19(16): p. R700-R713.
523. Saha, D., et al., Explosive sensing with insect-based biorobots. Biosensors and
Bioelectronics: X, 2020. 6: p. 100050.
524. Kuehn, M., et al., Changes of pressure and humidity affect olfactory function. European
archives of oto-rhino-laryngology, 2008. 265(3): p. 299-302.
525. Altundağ, A., et al., The effect of high altitude on olfactory functions. European Archives
of Oto-Rhino-Laryngology, 2014. 271(3): p. 615-618.
526. Martin, F., J. Riveron, and E. Alcorta, Environmental temperature modulates olfactory
reception in Drosophila melanogaster. Journal of insect physiology, 2011. 57(12): p.
1631-1642.
527. Stone, H., Influence of temperature on olfactory sensitivity. Journal of Applied
Physiology, 1963. 18(4): p. 746-751.
528. DeBerardinis, R.J. and C.B. Thompson, Cellular metabolism and disease: what do
metabolic outliers teach us? Cell, 2012. 148(6): p. 1132-1144.
529. Navale, A.M. and A.N. Paranjape, Glucose transporters: physiological and pathological
roles. Biophysical reviews, 2016. 8(1): p. 5-9.
530. Phypers, B. and J.T. Pierce, Lactate physiology in health and disease. Continuing
Education in Anaesthesia, Critical Care and Pain, 2006. 6(3): p. 128-132.
189
531. Rosenthal, M.D. and R.H. Glew, Medical biochemistry: human metabolism in health and
disease. 2011: John Wiley & Sons.
532. Association, A.D., Diagnosis and classification of diabetes mellitus. Diabetes care, 2009.
32(Supplement_1): p. S62-S67.
533. King, G.L. and M.R. Loeken, Hyperglycemia-induced oxidative stress in diabetic
complications. Histochemistry and cell biology, 2004. 122(4): p. 333-338.
534. Rolo, A.P. and C.M. Palmeira, Diabetes and mitochondrial function: role of
hyperglycemia and oxidative stress. Toxicology and applied pharmacology, 2006. 212(2):
p. 167-178.
535. Vanessa Fiorentino, T., et al., Hyperglycemia-induced oxidative stress and its role in
diabetes mellitus related cardiovascular diseases. Current pharmaceutical design, 2013.
19(32): p. 5695-5703.
536. Fernández-Real, J.M., A. López-Bermejo, and W. Ricart, Cross-talk between iron
metabolism and diabetes. Diabetes, 2002. 51(8): p. 2348-2354.
537. Anderson, J.C., A.L. Babb, and M.P. Hlastala, Modeling soluble gas exchange in the
airways and alveoli. Annals of biomedical engineering, 2003. 31(11): p. 1402-1422.
538. Sulway, M. and J. Malins, Acetone in diabetic ketoacidosis. The lancet, 1970. 296(7676):
p. 736-740.
539. Reichard Jr, G., et al., Acetone metabolism in humans during diabetic ketoacidosis.
Diabetes, 1986. 35(6): p. 668-674.
540. Galassetti, P.R., et al., Breath ethanol and acetone as indicators of serum glucose levels:
an initial report. Diabetes technology & therapeutics, 2005. 7(1): p. 115-123.
541. Wang, W., et al., Accuracy of breath test for diabetes mellitus diagnosis: a systematic
review and meta-analysis. BMJ Open Diabetes Research and Care, 2021. 9(1): p.
e002174.
542. Yatscoff, R.W., et al., 13C glucose breath test for the diagnosis of diabetic indications
and monitoring glycemic control. 2006, Google Patents.
543. Zhou, M.-G., et al., Investigation and identification of breath acetone as a potential
biomarker for type 2 diabetes diagnosis. Chinese Science Bulletin, 2014. 59(17): p. 1992-
1998.
544. Yan, K., et al., Design of a breath analysis system for diabetes screening and blood
glucose level prediction. IEEE transactions on biomedical engineering, 2014. 61(11): p.
2787-2795.
190
545. Turner, C., et al., Breath acetone concentration decreases with blood glucose
concentration in type I diabetes mellitus patients during hypoglycaemic clamps. Journal
of breath research, 2009. 3(4): p. 046004.
546. Righettoni, M. and A. Tricoli, Toward portable breath acetone analysis for diabetes
detection. Journal of breath research, 2011. 5(3): p. 037109.
547. Rydosz, A., Sensors for enhanced detection of acetone as a potential tool for noninvasive
diabetes monitoring. Sensors, 2018. 18(7): p. 2298.
548. Saasa, V., et al., Sensing technologies for detection of acetone in human breath for
diabetes diagnosis and monitoring. Diagnostics, 2018. 8(1): p. 12.
549. Minh, T.D.C., D.R. Blake, and P.R. Galassetti, The clinical potential of exhaled breath
analysis for diabetes mellitus. Diabetes research and clinical practice, 2012. 97(2): p.
195-205.
550. Warburg, O., On the origin of cancer cells. Science, 1956. 123(3191): p. 309-314.
551. Chen Jr, Y., et al., Lactate metabolism is associated with mammalian mitochondria.
Nature chemical biology, 2016. 12(11): p. 937-943.
552. de la Cruz-López, K.G., et al., Lactate in the regulation of tumor microenvironment and
therapeutic approaches. Frontiers in oncology, 2019. 9: p. 1143.
553. DeBerardinis, R.J., et al., The biology of cancer: metabolic reprogramming fuels cell
growth and proliferation. Cell metabolism, 2008. 7(1): p. 11-20.
554. Hirayama, A., et al., Quantitative metabolome profiling of colon and stomach cancer
microenvironment by capillary electrophoresis time-of-flight mass spectrometry. Cancer
research, 2009. 69(11): p. 4918-4925.
555. Gottfried, E., et al., Tumor-derived lactic acid modulates dendritic cell activation and
antigen expression. Blood, 2006. 107(5): p. 2013-2021.
556. Fischer, K., et al., Inhibitory effect of tumor cell–derived lactic acid on human T cells.
Blood, 2007. 109(9): p. 3812-3819.
557. Hanahan, D. and R.A. Weinberg, Hallmarks of cancer: the next generation. cell, 2011.
144(5): p. 646-674.
558. Fuchs, P., et al., Breath gas aldehydes as biomarkers of lung cancer. International
Journal of Cancer, 2010. 126(11): p. 2663-2670.
191
559. Leitch, B. and G. Laurent, GABAergic synapses in the antennal lobe and mushroom body
of the locust olfactory system. Journal of comparative Neurology, 1996. 372(4): p. 487-
514.
560. Laurent, G., Olfactory network dynamics and the coding of multidimensional signals.
Nature reviews neuroscience, 2002. 3(11): p. 884-895.
561. Bazhenov, M., et al., Model of transient oscillatory synchronization in the locust antennal
lobe. Neuron, 2001. 30(2): p. 553-567.
562. Sejnowski, T.J. and O. Paulsen, Network oscillations: emerging computational principles.
Journal of Neuroscience, 2006. 26(6): p. 1673-1676.
563. Turner, G.C., M. Bazhenov, and G. Laurent, Olfactory representations by Drosophila
mushroom body neurons. Journal of neurophysiology, 2008. 99(2): p. 734-746.
564. Ernst, K., J. Boeckh, and V. Boeckh, A neuroanatomical study on the organization of the
central antennal pathways in insects. Cell and tissue research, 1977. 176(3): p. 285-308.
565. Bergmann, G.A. and G. Bicker, Cholinergic calcium responses in cultured antennal lobe
neurons of the migratory locust. Scientific reports, 2021. 11(1): p. 1-15.
566. Stopfer, M., Central processing in the mushroom bodies. Current opinion in insect
science, 2014. 6: p. 99-103.
567. Jortner, R.A., S.S. Farivar, and G. Laurent, A simple connectivity scheme for sparse coding
in an olfactory system. Journal of Neuroscience, 2007. 27(7): p. 1659-1669.
568. Gupta, N. and M. Stopfer, A temporal channel for information in sparse sensory coding.
Current Biology, 2014. 24(19): p. 2247-2256.
569. Hanna, G.B., et al., Accuracy and methodologic challenges of volatile organic compound–
based exhaled breath tests for cancer diagnosis: a systematic review and meta-analysis.
JAMA oncology, 2019. 5(1): p. e182815-e182815.
570. Raman, B., et al., Temporally diverse firing patterns in olfactory receptor neurons
underlie spatiotemporal neural codes for odors. Journal of Neuroscience, 2010. 30(6): p.
1994-2006.
571. Goodwin, M.L., et al., Lactate and cancer: revisiting the Warburg effect in an era of
lactate shuttling. Frontiers in nutrition, 2015. 1: p. 27.
572. Boland, M.L., A.H. Chourasia, and K.F. Macleod, Mitochondrial dysfunction in cancer.
Frontiers in oncology, 2013. 3: p. 292.
192
573. Sabharwal, S.S. and P.T. Schumacker, Mitochondrial ROS in cancer: initiators, amplifiers
or an Achilles' heel? Nature Reviews Cancer, 2014. 14(11): p. 709-721.
574. Denko, N.C., Hypoxia, HIF1 and glucose metabolism in the solid tumour. Nature Reviews
Cancer, 2008. 8(9): p. 705-713.
575. DeBerardinis, R.J., Is cancer a disease of abnormal cellular metabolism? New angles on
an old idea. Genetics in Medicine, 2008. 10(11): p. 767-777.
576. Levine, A.J. and A.M. Puzio-Kuter, The control of the metabolic switch in cancers by
oncogenes and tumor suppressor genes. Science, 2010. 330(6009): p. 1340-1344.
577. Pfeiffer, T., S. Schuster, and S. Bonhoeffer, Cooperation and competition in the evolution
of ATP-producing pathways. Science, 2001. 292(5516): p. 504-507.
578. Shestov, A.A., et al., Quantitative determinants of aerobic glycolysis identify flux through
the enzyme GAPDH as a limiting step. elife, 2014. 3: p. e03342.
579. Vaupel, P., H. Schmidberger, and A. Mayer, The Warburg effect: essential part of
metabolic reprogramming and central contributor to cancer progression. International
journal of radiation biology, 2019. 95(7): p. 912-919.
580. Walenta, S., et al., High lactate levels predict likelihood of metastases, tumor recurrence,
and restricted patient survival in human cervical cancers. Cancer research, 2000. 60(4):
p. 916-921.
581. Sonveaux, P., et al., Targeting lactate-fueled respiration selectively kills hypoxic tumor
cells in mice. The Journal of clinical investigation, 2008. 118(12): p. 3930-3942.
582. Zent, R. and A. Pozzi, Cell-extracellular matrix interactions in cancer. 2010: Springer.
583. Hakim, M., et al., Volatile organic compounds of lung cancer and possible biochemical
pathways. Chemical reviews, 2012. 112(11): p. 5949-5966.
584. Gessner, C., et al., Detection of p53 gene mutations in exhaled breath condensate of
non-small cell lung cancer patients. Lung cancer, 2004. 43(2): p. 215-222.
585. Brunner, C., et al., Discrimination of cancerous and non-cancerous cell lines by
headspace-analysis with PTR-MS. Analytical and bioanalytical chemistry, 2010. 397(6): p.
2315-2324.
586. Sulé-Suso, J., et al., Quantification of acetaldehyde and carbon dioxide in the headspace
of malignant and non-malignant lung cells in vitro by SIFT-MS. Analyst, 2009. 134(12): p.
2419-2425.
193
587. Smith, D., et al., Quantification of acetaldehyde released by lung cancer cells in vitro
using selected ion flow tube mass spectrometry. Rapid communications in mass
spectrometry, 2003. 17(8): p. 845-850.
588. Elmassry, M.M. and M.A. Farag, In vivo and in vitro volatile organic compounds (VOCs)
analysis in bacterial diagnostics: case studies in agriculture and human diseases, in
Bacterial Volatile Compounds as Mediators of Airborne Interactions. 2020, Springer. p.
123-138.
589. Pereira, J., et al., Breath analysis as a potential and non-invasive frontier in disease
diagnosis: an overview. Metabolites, 2015. 5(1): p. 3-55.
590. Saha, D., et al., A spatiotemporal coding mechanism for background-invariant odor
recognition. Nature Neuroscience, 2013. 16(12): p. 1830-1839.
591. Pouzat, C., O. Mazor, and G. Laurent, Using noise signature to optimize spike-sorting and
to assess neuronal classification quality. Journal of Neuroscience Methods, 2002. 122(1):
p. 43-57.
592. Nizampatnam, S., et al., Dynamic contrast enhancement and flexible odor codes. Nat
Commun, 2018. 9(1): p. 3062.
593. Stopfer, M. and G. Laurent, Short-term memory in olfactory network dynamics. Nature,
1999. 402(6762): p. 664-8.
594. Perez-Orive, J., et al., Oscillations and Sparsening of Odor Representations in the
Mushroom Body. Science, 2002. 297(5580): p. 359.
595. Stopfer, M., et al., Impaired odour discrimination on desynchronization of odour-
encoding neural assemblies. Nature, 1997. 390: p. 70-74.
596. Saha, D., et al., Behavioural correlates of combinatorial versus temporal features of
odour codes. Nature Communications, 2015. 6(1): p. 6953.
597. Kaloumenou, M., et al., Breath Analysis: A Promising Tool for Disease Diagnosis-The Role
of Sensors. Sensors (Basel), 2022. 22(3).
598. Gouzerh, F., et al., Odors and cancer: Current status and future directions. Biochim
Biophys Acta Rev Cancer, 2022. 1877(1): p. 188644.
599. Brunner, C., et al., Discrimination of cancerous and non-cancerous cell lines by
headspace-analysis with PTR-MS. Analytical and Bioanalytical Chemistry, 2010. 397(6):
p. 2315-2324.
600. Altomare, D.F., et al., Exhaled volatile organic compounds identify patients with
colorectal cancer. British Journal of Surgery, 2013. 100(1): p. 144-150.
194
601. Altomare, D.F., et al., Chemical signature of colorectal cancer: case–control study for
profiling the breath print. BJS Open, 2020. 4(6): p. 1189-1199.
602. Amal, H., et al., Detection of precancerous gastric lesions and gastric cancer through
exhaled breath. Gut, 2016. 65(3): p. 400.
603. Amal, H., et al., Assessment of ovarian cancer conditions from exhaled breath.
International Journal of Cancer, 2015. 136(6): p. E614-E622.
604. Gruber, M., et al., Analysis of exhaled breath for diagnosing head and neck squamous
cell carcinoma: a feasibility study. British Journal of Cancer, 2014. 111(4): p. 790-798.
605. Mochalski, P., et al., Ex vivoemission of volatile organic compounds from gastric
cancer and non-cancerous tissue. Journal of Breath Research, 2018. 12(4): p. 046005.
606. Nardi-Agmon, I. and N. Peled, Exhaled breath analysis for the early detection of lung
cancer: recent developments and future prospects. Lung Cancer (Auckl), 2017. 8: p. 31-
38.
607. Phillips, M., et al., Rapid Point-Of-Care Breath Test Predicts Breast Cancer And Abnormal
Mammograms in Symptomatic Women. medRxiv, 2020: p. 2020.04.07.20042895.
608. Saidi, T., et al., Non-invasive prediction of lung cancer histological types through exhaled
breath analysis by UV-irradiated electronic nose and GC/QTOF/MS. Sensors and
Actuators B: Chemical, 2020. 311: p. 127932.
609. Wang, C., et al., Volatile Organic Metabolites Identify Patients with Breast Cancer,
Cyclomastopathy and Mammary Gland Fibroma. Scientific Reports, 2014. 4(1): p. 5383.
610. Španěl, P. and D. Smith, Quantification of volatile metabolites in exhaled breath by
selected ion flow tube mass spectrometry, SIFT-MS. Clin Mass Spectrom, 2020. 16: p. 18-
24.
611. Chen, H., et al., COVID-19 screening using breath-borne volatile organic compounds.
Journal of Breath Research, 2021.
612. Tiele, A., et al., Breath-based non-invasive diagnosis of Alzheimer’s disease: a pilot study.
Journal of Breath Research, 2020. 14(2): p. 026003.
613. Arasaradnam, R.P., et al., Detection of Colorectal Cancer (CRC) by Urinary Volatile
Organic Compound Analysis. PLOS ONE, 2014. 9(9): p. e108750.
614. Arasaradnam, R.P., et al., Noninvasive Diagnosis of Pancreatic Cancer Through
Detection of Volatile Organic Compounds in Urine. Gastroenterology, 2018.
154(3): p. 485-487.e1.
195
615. Koureas, M., et al., Target Analysis of Volatile Organic Compounds in Exhaled Breath for
Lung Cancer Discrimination from Other Pulmonary Diseases and Healthy Persons.
Metabolites, 2020. 10(8).
616. Tiele, A., et al., Breath Analysis Using eNose and Ion Mobility Technology to Diagnose
Inflammatory Bowel Disease-A Pilot Study. Biosensors (Basel), 2019. 9(2).
617. Baldini, C., et al., Electronic Nose as a Novel Method for Diagnosing Cancer: A Systematic
Review. Biosensors (Basel), 2020. 10(8).
618. Röck, F., N. Barsan, and U. Weimar, Electronic Nose: Current Status and Future Trends.
Chemical Reviews, 2008. 108(2): p. 705-725.
619. Broza, Y.Y., et al., Screening for gastric cancer using exhaled breath samples. British
Journal of Surgery, 2019. 106(9): p. 1122-1125.
620. Chang, J.-E., et al., Analysis of volatile organic compounds in exhaled breath for lung
cancer diagnosis using a sensor system. Sensors and Actuators B: Chemical, 2018. 255:
p. 800-807.
621. Díaz de León-Martínez, L., et al., Identification of profiles of volatile organic compounds
in exhaled breath by means of an electronic nose as a proposal for a screening method
for breast cancer: a case-control study. Journal of Breath Research, 2020. 14(4): p.
046009.
622. Kononov, A., et al., Online breath analysis using metal oxide semiconductor sensors
(electronic nose) for diagnosis of lung cancer. Journal of Breath Research, 2019. 14(1): p.
016004.
623. Krauss, E., et al., Recognition of breathprints of lung cancer and chronic obstructive
pulmonary disease using the Aeonose® electronic nose. Journal of Breath
Research, 2020. 14(4): p. 046004.
624. Roine, A., et al., Detection of Prostate Cancer by an Electronic Nose: A Proof of Principle
Study. The Journal of Urology, 2014. 192(1): p. 230-235.
625. van de Goor, R.M.G.E., et al., Feasibility of electronic nose technology for discriminating
between head and neck, bladder, and colon carcinomas. European Archives of Oto-
Rhino-Laryngology, 2017. 274(2): p. 1053-1060.
626. van Keulen, K.E., et al., Volatile organic compounds in breath can serve as a non-invasive
diagnostic biomarker for the detection of advanced adenomas and colorectal cancer.
Alimentary Pharmacology & Therapeutics, 2020. 51(3): p. 334-346.
627. Waltman, C.G., T.A.T. Marcelissen, and J.G.H. van Roermund, Exhaled-breath Testing for
Prostate Cancer Based on Volatile Organic Compound Profiling Using an Electronic Nose
196
Device (Aeonose™): A Preliminary Report. European Urology Focus, 2020. 6(6): p. 1220-
1225.
628. Barash, O., et al., Differentiation between genetic mutations of breast cancer by breath
volatolomics. Oncotarget, 2015. 6(42): p. 44864-76.
629. Chen, Y., et al., Breath Analysis Based on Surface-Enhanced Raman Scattering Sensors
Distinguishes Early and Advanced Gastric Cancer Patients from Healthy Persons. ACS
Nano, 2016. 10(9): p. 8169-8179.
630. Peng, G., et al., Detection of lung, breast, colorectal, and prostate cancers from exhaled
breath using a single array of nanosensors. British Journal of Cancer, 2010. 103(4): p.
542-551.
631. Raspagliesi, F., et al., Detection of Ovarian Cancer through Exhaled Breath by Electronic
Nose: A Prospective Study. Cancers (Basel), 2020. 12(9).
632. Karakaya, D., O. Ulucan, and M. Turkan, Electronic Nose and Its Applications: A Survey.
International Journal of Automation and Computing, 2020. 17(2): p. 179-209.
633. Covington, J., et al., The application of FAIMS gas analysis in Medical Diagnostics.
Analyst, 2015. 140.
634. Brooks, S.W., et al., Canine olfaction and electronic nose detection of volatile organic
compounds in the detection of cancer: a review. Cancer investigation, 2015. 33(9): p.
411-419.
635. Hurot, C., et al., Bio-Inspired Strategies for Improving the Selectivity and Sensitivity of
Artificial Noses: A Review. Sensors, 2020. 20(6).
636. Feng, X., et al., A living cell-based biosensor utilizing G-protein coupled receptors:
principles and detection methods. Biosens Bioelectron, 2007. 22(12): p. 3230-7.
637. Ko, H.J. and T.H. Park, Bioelectronic nose and its application to smell visualization. J Biol
Eng, 2016. 10: p. 17.
638. Lee, S.H., E.H. Oh, and T.H. Park, Cell-based microfluidic platform for mimicking human
olfactory system. Biosens Bioelectron, 2015. 74: p. 554-61.
639. Hwi Jin, K. and P. Tai Hyun, Dual signal transduction mediated by a single type of
olfactory receptor expressed in a heterologous system. Biological Chemistry, 2006.
387(1): p. 59-68.
640. Lee, S.H., et al., Cell-based olfactory biosensor using microfabricated planar electrode.
Biosensors and Bioelectronics, 2009. 24(8): p. 2659-2664.
197
641. Lee, S.H., et al., Enhancement of cellular olfactory signal by electrical stimulation.
ELECTROPHORESIS, 2009. 30(18): p. 3283-3288.
642. Misawa, N., et al., Construction of a Biohybrid Odorant Sensor Using Biological Olfactory
Receptors Embedded into Bilayer Lipid Membrane on a Chip. ACS Sens, 2019. 4(3): p.
711-716.
643. Mitsuno, H., et al., Novel cell-based odorant sensor elements based on insect odorant
receptors. Biosens Bioelectron, 2015. 65: p. 287-94.
644. Termtanasombat, M., et al., Cell-Based Odorant Sensor Array for Odor Discrimination
Based on Insect Odorant Receptors. J Chem Ecol, 2016. 42(7): p. 716-24.
645. Murugathas, T., et al., Biosensing with Insect Odorant Receptor Nanodiscs and Carbon
Nanotube Field-Effect Transistors. ACS Appl Mater Interfaces, 2019. 11(9): p. 9530-9538.
646. Sato, K. and S. Takeuchi, Chemical Vapor Detection Using a Reconstituted Insect
Olfactory Receptor Complex. Angewandte Chemie International Edition, 2014. 53(44): p.
11798-11802.
647. Gardner, E.P. and J.H. Martin, Coding of sensory information. Principles of neural
science, 2000. 4: p. 411-429.
648. Malnic, B., et al., Combinatorial receptor codes for odors. Cell, 1999. 96(5): p. 713-723.
649. Kandel, E.R., et al., Principles of neural science. Vol. 4. 2000: McGraw-hill New York.
650. Moran, J.K., et al., The scent of blood: a driver of human behavior? PLoS One, 2015.
10(9): p. e0137777.
651. Sarrafchi, A. and M. Laska, Olfactory sensitivity for the mammalian blood odor
component trans-4, 5-epoxy-(E)-2-decenal in CD-1 mice. Perception, 2017. 46(3-4): p.
333-342.
652. Giurfa, M. and J.-C. Sandoz, Invertebrate learning and memory: fifty years of olfactory
conditioning of the proboscis extension response in honeybees. Learning & memory,
2012. 19(2): p. 54-66.
653. Wenner, A.M., P.H. Wells, and D.L. Johnson, Honey bee recruitment to food sources:
olfaction or language? Science, 1969. 164(3875): p. 84-86.
654. Galizia, C.G., S.L. McIlwrath, and R. Menzel, A digital three-dimensional atlas of the
honeybee antennal lobe based on optical sections acquired by confocal microscopy. Cell
and tissue research, 1999. 295(3): p. 383-394.
198
655. Flanagan, D. and A.R. Mercer, An atlas and 3-D reconstruction of the antennal lobes in
the worker honey bee, Apis mellifera L.(Hymenoptera: Apidae). International Journal of
Insect Morphology and Embryology, 1989. 18(2-3): p. 145-159.
656. Brandt, R., et al., Three-dimensional average-shape atlas of the honeybee brain and its
applications. Journal of Comparative Neurology, 2005. 492(1): p. 1-19.
657. Rybak, J., The digital honey bee brain atlas, in Honeybee neurobiology and behavior.
2012, Springer. p. 125-140.
658. Sachse, S. and C.G. Galizia, The coding of odour-intensity in the honeybee antennal lobe:
local computation optimizes odour representation. European journal of neuroscience,
2003. 18(8): p. 2119-2132.
659. Galizia, C.G. and R. Menzel, The role of glomeruli in the neural representation of odours:
results from optical recording studies. Journal of insect physiology, 2001. 47(2): p. 115-
130.
660. Fialho, M.d.C.Q., et al., A comparative study of the antennal sensilla in corbiculate bees.
Journal of Apicultural Research, 2014. 53(3): p. 392-403.
661. Esslen, J. and K.-E. Kaissling, Zahl und Verteilung antennaler Sensillen bei der Honigbiene
(Apis mellifera L.). Zoomorphologie, 1976. 83(3): p. 227-251.
662. Forêt, S. and R. Maleszka, Function and evolution of a gene family encoding odorant
binding-like proteins in a social insect, the honey bee (Apis mellifera). Genome research,
2006. 16(11): p. 1404-1413.
663. Robertson, H.M. and K.W. Wanner, The chemoreceptor superfamily in the honey bee,
Apis mellifera: expansion of the odorant, but not gustatory, receptor family. Genome
research, 2006. 16(11): p. 1395-1403.
664. Akers, R.P. and W.M. Getz, A test of identified response classes among olfactory receptor
neurons in the honey-bee worker. Chemical Senses, 1992. 17(2): p. 191-209.
665. Suzuki, H., Convergence of olfactory inputs from both antennae in the brain of the
honeybee. 1975.
666. Sinakevitch, I.T., et al., Apis mellifera octopamine receptor 1 (AmOA1) expression in
antennal lobe networks of the honey bee (Apis mellifera) and fruit fly (Drosophila
melanogaster). Frontiers in systems neuroscience, 2013. 7: p. 70.
667. Galizia, C.G. and P. Szyszka, Olfactory coding in the insect brain: molecular receptive
ranges, spatial and temporal coding. Entomologia experimentalis et applicata, 2008.
128(1): p. 81-92.
199
668. Brodie, A. and N. Vasdev, The future of robotic surgery. The Annals of The Royal College
of Surgeons of England, 2018. 100(Supplement 7): p. 4-13.
669. Hanson, T.L., et al., The “sewing machine” for minimally invasive neural recording.
bioRxiv, 2019: p. 578542.
670. Musk, E., An Integrated Brain-Machine Interface Platform With Thousands of Channels. J
Med Internet Res, 2019. 21(10): p. e16194.
671. Salatino, J.W., et al., Glial responses to implanted electrodes in the brain. Nature
biomedical engineering, 2017. 1(11): p. 862-877.
672. Steinmetz, N.A., et al., Neuropixels 2.0: A miniaturized high-density probe for stable,
long-term brain recordings. Science, 2021. 372(6539): p. eabf4588.
673. Obaid, A., et al., Massively parallel microwire arrays integrated with CMOS chips for
neural recording. Science Advances. 6(12): p. eaay2789.
674. Hashemi Noshahr, F., M. Nabavi, and M. Sawan, Multi-Channel Neural Recording
Implants: A Review. Sensors (Basel), 2020. 20(3).
675. Schmidt, K. and I. Podmore, Current Challenges in Volatile Organic Compounds Analysis
as Potential Biomarkers of Cancer. Journal of Biomarkers, 2015. 2015: p. 981458.
676. Jia, Z., et al., Critical Review of Volatile Organic Compound Analysis in Breath and In Vitro
Cell Culture for Detection of Lung Cancer. Metabolites, 2019. 9(3).
200